Fourier Transform Applications

FOURIER TRANSFORM
APPLICATIONS

Edited by Salih Mohammed Salih

Fourier Transform Applications
Edited by Salih Mohammed Salih

Published by InTech
Janeza Trdine 9, 51000 Rijeka, Croatia

Copyright 2012 InTech
All chapters are Open Access distributed under the Creative Commons Attribution 3.0
license, which allows users to download, copy and build upon published articles even for
commercial purposes, as long as the author and publisher are properly credited, which
ensures maximum dissemination and a wider impact of our publications. After this work
has been published by InTech, authors have the right to republish it, in whole or part, in
any publication of which they are the author, and to make other personal use of the
work. Any republication, referencing or personal use of the work must explicitly identify
the original source.

As for readers, this license allows users to download, copy and build upon published
chapters even for commercial purposes, as long as the author and publisher are properly
credited, which ensures maximum dissemination and a wider impact of our publications.

Notice
Statements and opinions expressed in the chapters are these of the individual contributors
and not necessarily those of the editors or publisher. No responsibility is accepted for the
accuracy of information contained in the published chapters. The publisher assumes no
responsibility for any damage or injury to persons or property arising out of the use of any
materials, instructions, methods or ideas contained in the book.

Publishing Process Manager Vana Persen
Technical Editor Teodora Smiljanic
Cover Designer InTech Design Team

First published April, 2012
Printed in Croatia

A free online edition of this book is available at www.intechopen.com
Additional hard copies can be obtained from orders@intechopen.com

Fourier Transform Applications, Edited by Salih Mohammed Salih
p. cm.
ISBN 978-953-51-0518-3

Contents

Preface IX
Part 1 Electromagnetic Field and Microwave Applications 1
Chapter 1 Computation of Transient Near-Field
Radiated by Electronic Devices from Frequency Data 3
Blaise Ravelo and Yang Liu
Chapter 2 Impulse-Regime Analysis of Novel
Optically-Inspired Phenomena at Microwaves 27
J. Sebastian Gomez-Diaz, Alejandro Alvarez-Melcon,
Shulabh Gupta and Christophe Caloz
Chapter 3 Fourier Transform Application in
the Computation of Lightning Electromagnetic Field 57
Vesna Javor
Chapter 4 Robust Beamforming and DOA Estimation 87
Liu Congfeng
Chapter 5 Analysis of Long-Periodic Fluctuations of
Solar Microwave Radiation, as a Way for
Diagnostics of Coronal Magnetic Loops Dynamics 143
Maxim L. Khodachenko, Albert G. Kislyakov and Eugeny I. Shkelev
Part 2 Medical Applications 167
Chapter 6 Spectral Analysis of Heart Rate Variability in Women 169
Ester da Silva, Ana Cristina S. Rebelo,
Nayara Y. Tamburs, Mariana R. Salviati,
Marcio Clementino S. Santos and Roberta S. Zuttin
Chapter 7 Cortical Specification of a Fast Fourier Transform
Supports a Convolution Model of Visual Perception 181
Phillip Sheridan
VI Contents

Chapter 8 Spectral Analysis of Global
Behaviour of C. Elegans Chromosomes 205
Afef Elloumi Oueslati, Imen Messaoudi,
Zied Lachiri and Noureddine Ellouze
Part 3 Fourier and Helbert Transform Applications 229
Chapter 9 The Fourier Convolution Theorem over Finite Fields:
Extensions of Its Application to Error Control Coding 231
Eric Sakk and Schinnel Small
Chapter 10 Application of the Weighted Energy Method in
the Partial Fourier Space to Linearized Viscous
Conservation Laws with Non-Convex Condition 249
Yoshihiro Ueda
Chapter 11 Fourier Transform Methods for Option Pricing 265
Deng Ding
Chapter 12 Hilbert Transform and Applications 291
Yi-Wen Liu

Preface

During the preparation of this book, we found that almost all the textbooks on signal
analysis have a section devoted to the Fourier transform theory. The Fourier
transform is a mathematical operation with many applications in physics, and
engineering that express a mathematical function of time as a function of frequency,
known as its frequency spectrum; Fourier's theorem guarantees that this can always
be done. The basic idea behind all those horrible looking formulas is rather simple,
even fascinating: it is possible to form any function as a summation of a series of sine and
cosine terms of increasing frequency. In other words, any space or time varying data can
be transformed into a different domain called the frequency space. A fellow called
Joseph Fourier first came up with the idea in the 19
th
century, and it was proven to be
useful in various applications, mainly in signal processing. As far as we can tell,
Gauss was the first to propose the techniques that we now call the Fast Fourier
Transform (FFT) for calculating the coefficients in a trigonometric expansion of an
asteroid's orbit in 1805. However, it was the seminal paper by Cooley and Tukey in
1965 that caught the attention of the science and engineering community and, in a
way, founded the discipline of digital signal processing. While the Discrete Fourier
Transform (DFT) can be applied to any complex valued series, in practice for large
series it can take considerable time to compute, the time taken being proportional to
the square of the number on points in the series. It is hard to overemphasis the
importance of the DFT, convolution, and fast algorithms. The FFT may be the most
important numerical algorithm in science, engineering, and applied mathematics.
New theoretical results are still appearing, advances in computers and hardware
continually restate the basic questions, and new applications open new areas for
research. It is hoped that this book will provide the background, references and
incentive to encourage further research and results in this area as well as provide
tools for practical applications. One of the attractive features of this book is the
inclusion of extensive simple, but practical, examples that expose the reader to real-
life signal analysis problems, which has been made possible by the use of computers
in solving practical design problems. The aim of this book is to expand the
applications of Fourier transform into main three sections:
The first section deals with Electromagnetic Field and Microwave Applications. It consists
of five chapters. The chapters are related to: the computation of transient near-field
radiated by electronic devices, analysis of novel optically-inspired phenomena at
X Preface

microwaves, the computation of lightning electromagnetic field, beamforming and
DOA estimation, and the solar microwave radiation.
The chapters of the second section discuss some advanced methods used in Fourier
transform analysis which are related to the Medical Applications. This section consists of
three chapters. The chapters are related to: spectral analysis of global behaviour of C.
Elegans chromosomes, spectral analysis of heart rate variability in women, and the
cortical specification of a fast Fourier transform supports a convolution model of
visual perception.
The third section includes the Fourier and Helbert Transform Applications. This section
consists of four chapters. The chapters are concerns to: the Fourier convolution
theorem over finite fields (error control coding), application of the weighted energy
method in the partial Fourier space to linearized viscous conservation laws with non-
convex condition, Fourier transform methods for option pricing, and the Hilbert
transform applications.
Finally, we would like to thank all the authors who have participated in this book for
their valuable contribution. Also we would like to thank all the reviewers for their
valuable notes. While there is no doubt that this book may have omitted some
significant findings in the Fourier transform field, we hope the information included
will be useful for electrical engineers, control engineers, communication engineers,
signal processing engineers, medical researchers, and the mathematicians, in addition
to the academic researchers working in the above fields.

Salih Mohammed Salih
College of Engineering
University of Anbar
Iraq

Part 1
Electromagnetic Field and
Microwave Applications

1
Computation of Transient Near-Field Radiated
by Electronic Devices from Frequency Data
Blaise Ravelo and Yang Liu
IRSEEM (Research Institute in Embedded Electronic System), EA 4353,
Graduate School of Engineering ESIGELEC,
76801 Saint Etienne du Rouvray Cedex,
France
1. Introduction
Facing to the increase of architecture complexity in the modern high-speed electronic
equipments, the electromagnetic compatibility (EMC) characterization becomes a crucial
step during the design process. This electromagnetic (EM) characterization can manifest
with the unintentional conducting or radiating perturbations including, in particular, the
near-field (NF) emissions. Accurate modelling method of this emission in NF zone becomes
one of electronic engineer designers and researchers most concerns (Shi et al. 1989, Baudry
et al. 2007, Vives-Gilabert et al. 2007, Vives-Gilabert et al. 2009, Song et al. 2010, Yang et al.
2010). This is why since the middle of 2000s; the NF modelling has been a novel speciality of
the electronic design engineers. This modelling technique enables a considerable insurance
of the reliability and the safety of the new electronic products. To avoid the doubtful issues
related to the EM coupling, this analysis seems indispensable for the modern RF/digital
electronic boards vis--vis the growth of the integration density and the operating numerical
data-speed which achieves nowadays several Gbit/s (Barriere et al. 2009, Archambeault et
al. 2010). In this scope, the influence of EM-NF-radiations in time-domain and in ultra-wide
band (UWB) RF-/microwave-frequencies remains an open-question for numerous electronic
researchers and engineer designers (Ravelo et al. 2011a & 2011b, Liu Y. et al. 2011a & 2011b).
In the complex structures, the current and voltage commutations in the non-linear electronic
devices such as diodes, MOSFETs and also the amplifiers can create critical undesired
transient perturbations (Jauregui et al. 2010a, Vye 2011, Trscher 2011, Kopp 2011). Such
electrical perturbations are susceptible to generate transient EM-field radiations which need
to be modelled and mastered by the electronic handset designers and manufacturers.
1.1 Overview on the NF radiations characterization occurring in the RF/microwave-
device in time-domain
It is noteworthy that the frequency-investigations on the EM-radiation of electronic devices
are not sufficient for the representation of certain EM-transient phenomena notably when
the sources of perturbations behave as a short duration pulse-wave. In fact, it does not
enable to precise the probably instant times and the intensity peak of the EM-pulse. That is
why the time-domain representation is particularly essential for the infrequent and ultra-


4
short duration wave emission analysis. In order to investigate more concretely the
unwanted time-domain perturbations, different EM-NF modelling and measurement
techniques were recently introduced and published in the literature (Cicchetti 1991, Adada
2007, Liu L. et al. 2009, Winter & Herbrig 2009, Ordas et al. 2009, Braun et al. 2009, Rioult et
al. 2009, Xie & Lei 2009, Edwards et al. 2010, Jauregui et al. 2010b, Ravelo 2010).
Furthermore, several EM-solvers are also integrated in the commercial simulation tools for
the determination of the EM-field radiations by the RF/microwave devices especially in
frequency domain (ANSOFT 2006, AGILENT 2008, ANSYS 2009, NESA 2010).
Currently, the computation method of the EM-field becomes systematically more and more
complicated when the electronic systems operate with baseband UWB signals. Despite the
recent investigations conducted on the finite-difference time-domain method (FDTD)
method (Liu et al. 2009, Jauregui et al. 2010b), the accuracy of the computation results with
these time-domain commercial tools remains difficult to evaluate when the perturbation
sources are induced from ultra-short duration transient NF. In addition, more practical
techniques (Cicchetti 1991, Braun et al. 2009, Winter & Herbrig 2009, Ordas 2009, Rioult et al.
2009) have been also introduced for the measurement of the electric- and electronic- system
electromagnetic interference (EMI). But compared to the existing frequency measurement
techniques, they are much better because of the limitations either in terms of space-
resolution or electro-sensitivity or simply the calibration process. So, the evaluation of the
accurate graphs of time-dependent EM-waves in NF is still an open challenge.
To cope with this limitation, in this chapter, an efficient computation methodology based on
the transformation of wide bandwidth and baseband frequency-dependent data for the
determination of the transient EM-NF mapping permitting is developed. In order to take
into account the transient radiations specific to the expected use cases, an adequate
excitation signal should be considered. This excitation is usually defined according to
certain technical parameters (amplitude, temporal width, variation speed, time-duration)
which qualifies the undesired disturbing signal susceptible to propagate in the emitting
circuits. Then, the fast Fourier transform (fft) mathematical treatment of the assumed
disturbing signal synchronized with the given discrete frequency-dependent data in the
adequate frequency range enables to determine the transient wave radiation mapping.
1.2 Background on the EMC application of the transient EM-NF
As aforementioned and discussed (Rammal et al. 2009, Jauregui et al. 2010a), the EM-transient
analysis is actually important for the immunity predictions in the mixed or analogue-digital
components constituting the high-speed electronic boards regarding the eventual radiations
of high power electrical circuitry as the case of neighbouring hybrid electric vehicle
propulsion systems. To assess such an EMC effect, as reported in (Adada 2007), the
electronic circuit designers working on analogue/mixed signal (AMS) subsystems have
preferred software tools such as SPICE, while those working on RF/microwave front-end
components have tended to manipulate S-parameter frequency-domain design and
simulation tools. By cons, currently, the fusion of the both approaches as AMS engineers are
required to make further analysis on the critical components is needed by using the
adequate EM simulation tools. In this case, we have to elaborate the context of ultra-wide
band (UWB). Currently, this topic is one of improvement techniques in the area of EMC
application. In this optic, the modelling of mixed component EM-NF emission becomes one

Computation of Transient Near-Field Radiated by Electronic Devices from Frequency Data

5
of the crucial steps before the implementation process. Therefore, the undesired EMC
radiations should be investigated not only in frequency-domain but also in time-domain.
For this reason, we propose in this chapter an extraction method enabling to determine the
time-domain EM-NF maps from the frequency-dependent data by using the Fourier
transform of the 2D data.
1.3 Outline of the presented chapter
To make this chapter better to understand, it is organized in three main sections. Section 2
describes the methodology of the time-frequency computation-method proposed. It details
how to extract the transient EM-NF radiation from the given time-dependent excitation
sampled signal and the frequency-dependent data. Then, more concrete validation of the
computation-method investigated by considering the EM-NF radiated by an arbitrary set of
magnetic dipoles is devoted in Section 3. The EM-NF reference data are calculated with the
theoretical formulas introduced in (Baum 1971 & 1976, Singaraju & Baum 1976). As
reviewed by certain research works (Hertz 1892, Chew & Kong 1981, Lakhtakiaa et al. 1987,
Song & Chen 1993, Jun-Hong et al 1997, Schantz 2001, Selin 2001, Smagin & Mazalov 2005,
Sten & Hujanen 2006, Ravelo 2010), the analytical calculation performed with the EM-wave
emitted by elementary dipoles allows to realize more practical and more explicit
mathematical analyses of the EM-field expressions in different physic areas. We point out
that the EM-field emitted by electronic devices can be modelled by the radiations of the
optimized combination of elementary EM-dipoles (Fernndez-Lpez et al. 2009). To confirm
the feasibility of the method proposed, an application with another proof of concept with a
concrete electronic device is also offered in Section 4. This practical verification will be made
toward a microwave electronic design of low-pass planar microstrip filter operating up until
some GHz. Lastly; Section 5 draws the conclusion of this chapter.
2. Methodology of the time-frequency computation method investigated
The present section is divided in two different parts. First, an explicit description illustrates
how to examine the transient excitation signal for the UWB applications. Afterward, the
development of the routine process indicating the algorithm of the computation method
proposed is elaborated.
2.1 Frequency coefficient extraction
Let us denote i(t) the transient current which is considered also as the excitation of the under
test electronic structure. The sampled data corresponding to this test signal is supposed
discretized from the starting time t
min
to the stop time t
max
with time step equal to t. In this
case, the number n of time-dependent samples is logically, equal to:

max min
int
t t
n
t

=

, (1)
with int(x) expresses the lowest integer number greater than the real x. Accordingly, via the
fast Fourier transform (fft), the equivalent frequency-dependent spectrum of i(t
k
) (with t
k
=
k.t and k = {1n}) can be determined. The frequency data emanated by this mathematical
transform are generally as a complex number denoted by [ ] ( ) ( )
k k
I f fft i t = . Therefore, the


6
magnitude of this signal spectrum from DC to a certain frequency in function of the initial
time-step sampled data can be extracted as depicted in Fig. 2. Consequently, by denoting I
0

the sinusoidal current magnitude for generating the magnetic field spectrum
0
( ) H f , the
following complex coefficients of the input current in function of frequency as illustrated in
Fig. 1:

0
( )
step
k
I k f
c
I
= . (2)

Fig. 1. Extraction of the frequency coefficients from the excitation signal spectrum
It corresponds to the discrete data at each sample of frequency f
k
= k.f
step
for k = {1n} having
a step-frequency given by:

max min
1
step
f
t t
=
. (3)
We underline that this method requires a frequency range [f
min
, f
max
] whose the lowest
frequency value f
min
of ( ) I f -data is equal to the step frequency f
step
. It means that the
spectrum value can be extrapolated linearly to generate the excitation signal steady-state
component at f = 0. In practice, it does not change the calculation results because according
to the signal processing theory, the DC-component of transient waves with ultra-short time
duration at very low frequencies is usually negligible. The upper frequency f
max
should
correspond to the frequency bandwidth containing 95-% of the excitation signal considered
spectrum energy.
2.2 Routine process of the computation-method proposed
The computation-method under study is mainly consisted of two different steps. The first
step is focused on the time-domain characterization of the excitation current considered i(t
k
)
in the specific interval range varying from t
min
to t
max
with step t. In this first step, the
complex frequency-coefficients
k
c should be extracted through the fft-operation as
explained in previous subsection. The following second step is the conversion of the


7
frequency-dependent magnetic- or H-field expressed by
0 0
( , , , ) H x y z f which is recorded at
the point
0
( , , ) M x y z chosen arbitrarily, into a time-dependent data denoted by
0
( , , , ) H x y z t
by using the ifft-operation. These points
0
( , , ) M x y z belong in the X-Y plane positioned at
z = z
0
. In this case, the frequency range considered varying from f
min
to f
max
and the frequency
step f
step
of
0 0 0
( ) ( , , , ) H f H x y z f = must be well-synchronized with that of the excitation
signal frequency coefficients c
k
. Under this condition, the time-dependent data desired
0
( ) ( , , , ) H t H x y z t = which is generated by the specific excitation signal i(t) can be calculated
with the inverse fast Fourier transform (ifft) of the convolution product between
0
( ) ( ) / c f I f I = and
0
( ) H f written as:
{ }
0
( ) [ ( ). ( )] H t e ifft c f H f = . (4)
with { } e x represents the real part of the complex number x. Fig. 2 depicts the flow work
highlighting the different operations to be fulfilled for the achievement of the transient EM-
field computation-method proposed. This method enables to provide the time-dependent

Fig. 2. Flow work illustrating the transient H-field radiation computation-method proposed
knowing the temporal range, t
min
and t
max
step t of the excitation signal i(t
k
) and also the
frequency-dependent H-field
0
( ) H f (here the under bar indicates the complex variables)


8
H-field here denoted as H(t) according to the arbitrary form of the excitation and also
knowing the frequency-dependent H-field data in the frequency range starting from the
lowest value to the upper frequency limit equal to the inverse of the time-step t of the
discrete data x(t
k
).
3. Validation with the radiation of a set of magnetic dipoles
This section presents the verification of the computation method established regarding the
radiation of magnetic dipoles shown in Fig. 3 in time-domain. The results were realized by
considering the frequency-dependent EM-NF-field directly calculated from predefined
analytical formulas (Baum 1971 & 1976, Singaraju & Baum 1976, Balanis 2005). After the
analytical description of the considered dipole source and also the mathematical definitions
of the H-field expressions in frequency- and time-domains, we will explore the numerical
computation with Matlab programming.

Fig. 3. Configuration of the elementary magnetic dipole formed by a circular loop with
radius a placed at the centre O(0,0,0) of the cartesian (x,y,z)-coordinate system
In this figure, the magnetic dipole assumed as a circular wire having radius a is positioned
at the origin of the system. Then the considered point M where the EM-field will be
evaluated can be referred either in cartesian coordinate (x,y,z) or in spherical coordinate
(r,,). By assuming that the magnetic loop depicted in Fig. 3 is fed by transient current
denoted i(t), we have the H-field expressions recalled in appendix A (Baum 1971 & 1976,
Singaraju & Baum 1976).
3.1 Description of the radiation source considered
Fig. 4 displays the arbitrary placement representing the set of eight magnetic dipoles in the
horizontal plane Oxy considered for the validation of the calculation method under study.
To generate the various behaviours of EM-field graphs, the dipole axes are randomly
oriented as follows: M
1
, M
4
, M
5
and M
8
along Ox-axis, M
2
and M
6
along Oy, and M
3
and M
7

along Oz. These dipoles are in this case, flowed by an ultra-short duration transient current
i(t) which is considered as a pulse signal. These elementary dipoles are supposed formed by
wire circular loops having radius a = 0.5 mm.


9
Note that in the present study, the elementary dipoles are supposed ideal, thus, there is no
coupling between each other. By reason of linearity, the total EM-field at any point M(x,y,z)
is consequently the sum of each dipole contribution:

8
1
( , , ) ( , , )
k
M
k
H x y z H x y z
=
=

JJG JJG
. (5)
After implementation of the computation algorithm introduced in Fig. 2 in Matlab program,
we obtain the EM-calculation results presented in the next subsection.

Fig. 4. Configuration of the radiating source considered which is comprised of magnetic
dipoles placed in the horizontal plane xy
3.2 Validation results
In this subsection, comparison results between the transient EM-field maps radiated by the
elementary dipoles displayed in Fig. 4 from the direct calculation and from the method
proposed are presented.
3.2.1 Description of the excitation signal
In order to highlight the influence of the form and the transient variation of the disturbing
currents in the electronic structure, the considered short-duration pulse excitation current
i(t) is assumed as a bi-exponential signal analytically defined in appendix B. One points out
that in order to take into account the truncation effect between t
min
and t
max
, the considered
sampling data from i(t) should be multiplied by a specific time gate. So that accordingly,
each component ( ) I should be assumed as a sine cardinal. But here, this effect can be
negligible if the assumed time step is well-accurate. The numerical application was made by
taking the current amplitude I
M
= 1A and the time-constants
1
=
2
/2 = 2 ns. So, from the
analytical relation expressed in (B-4), we have
95%
3.07 Grad.s
-1
. Fig. 5 displays the
transient plot of this current excitation.
The time interval range of signal test was defined from t
min
= 0 ns to t
max
= 20 ns with step
t = 0.2 ns. One can see that this baseband signal presents a frequency bandwidth f
max
of
about 2 GHz, where belongs more than 95-% of the spectrum signal energy. The data
calculated ( ) [ ( )] I fft i t = generates the frequency-coefficient values of i(t) according to the
relation expressed in (2) as described earlier in subsection 2.2.


10
3.2.2 Discussions on the computed results
By considering the set of eight magnetic dipoles presented in Fig. 4 which are excited by the
same pulse current plotted in Fig. 5 yields the H-field component (H
x
, H
y
and H
z
) mappings
depicted in Fig. 6 at the arbitrary time t
0
= 2 ns and in the horizontal plane parallel to (Oxy)
referenced z
0
= 6.5 mm above the radiating source.

Fig. 5. Transient plot of the considered excitation current i(t)

Fig. 6. Maps of H-field components detected at the height z
0
= 6.5 mm above the radiating
source: (a) H
x
, (b) H
y
and (c) H
z
obtained from the direct calculation


11
This height was arbitrarily chosen in order to generate a significant NF effect in the
considered frequency range. The dimensions of the mapping plane were set at L
x
= 110 mm
and L
y
= 100 mm with space-resolution equal to x = y = 2 mm. First, by using the
harmonic expressions of the magnetic field components, the maps of the frequency-
dependent H-field are obtained from f
min
= 0.05 GHz to f
max
= 2.50 GHz with step f = 0.05
GHz. Fig. 7 represents the corresponding mappings of the H-field component magnitudes at
f
0
= 2 GHz.

Fig. 7. Maps of H-field components magnitude obtained at the frequency f
0
= 2 GHz: (a) H
x
,
(b) H
y
and (c) H
z
directly calculated from expressions (A-8), (A-9) and (A-10)
After the Matlab program implementation of the algorithm indicated by the flow chart
schematized in Fig. 2, the results shown in Figs. 8(a)-(d) are obtained via the combination of
the frequency-dependent data of the H-field components associated to the frequency-
coefficients of the excitation signal plotted in Fig. 5. One can see that the EM-maps
presenting the same behaviors as those obtained via the direct calculations displayed in Fig.
6 were established. In addition, we compare also as illustrated in Fig. 9 the modulus of the
H-fields from the method under study and from the 3D EM Field Simulator - CST
(Computer Simulation Technology). Furthermore, as evidenced by Figs. 10(a)-(c), very good
correlation between the profiles of the H-field components detected in the vertical cut-plane
along Oy and localized at x = 23 mm was observed. To get further insight about the time-
dependent representation of the H-field components, curves showing the variations of H
x
(t),
H
y
(t) and H
z
(t) at the arbitrary point chosen of the mapping plane having coordinates (x = 19
mm, y = 35 mm) are plotted in Figs. 11(a)-(c).


12
As results, once again, we can find that the H-field components from the frequency data fit
very well the direct calculated ones. As aforementioned, due to the truncation effects, the
H
x
-component presents a slight divergence at the ending time of the signal. This is
particularly due to the numerical noises at the very low value of the EM field as the case of
the x-component which is absolutely twenty times less than the two other components.
In order to prove in more realistic way the relevance of the investigated method, one
proposes to treat the radiation of concrete electronic devices in the next section.

Fig. 8. Maps of H-field components calculated from the time-frequency computation method
proposed for z
0
= 6.5 mm: (a) H
x
, (b) H
y
, (c) H
z

Fig. 9. Comparison of H-field maps modulus |H|(t
0
= 2 ns) obtained from the direct
formulae (in left) and from the time-frequency computation method proposed for
z
0
= 6.5 mm


13

Fig. 10. Comparisons of the H-field component profiles obtained from the time-frequency
computation method proposed and the direct calculation, detected in the vertical cut-plane
x = 23 mm: (a) H
x
, (b) H
y
and (c) H
z


14

Fig. 11. Comparisons of the H-field components temporal variation obtained from the time-
frequency computation method proposed and the direct calculation, detected in the arbitrary
point (x = 19 mm, y = 35 mm): (a) H
x
, (b) H
y
and (c) H
z


15
4. Application with the Transient NF emitted by a microstrip device proof-of-
concept
To get further insight about the feasibility of the computation method under study, let us
examine the transient EM-wave emitted by an example of more realistic microwave device.
This latter was designed with the standard 3-D EM-tools HFSS for generating the frequency-
dependent data which is used for the determination of transient H-NF. Then, the simulation
with CST microwave studio (MWS) was performed for the computation of the reference
H-NF mappings in time-domain. As realistic and concrete demonstrator, a low-pass
Tchebychev filter implemented in planar microstrip technology was designed. Its layout top
view including the geometrical dimensions is represented in Fig. 12(a).
This device was printed on the FR4-epoxy substrate having relativity permittivity
r
= 4.4,
thickness h = 1.6 mm and etched Cu-metal thickness t = 35 m. The cut cross section of this
microwave circuit is pictured in Fig. 12(b).

(a)

(b)
Fig. 12. (a) Top view of the CST design of the under test low-pass microstrip filter (b) Cross-
section cut of the under test microstrip filter with metallization thickness t = 35 m and
dielectric substrate height h = 1.6 mm
After simulations, we realize the results obtained and discussed in next subsections.


16
4.1 CST-computation results
The low-pass filter under test was simulated with CST MWS in the time interval range
delimited by t
min
= 0 ns and t
max
= 20 ns with step t = 0.2 ns. Therefore, the magnetic-field
maps are presented in Fig. 13. These curves are recorded at the arbitrary instant time t
0
= 2
ns. Note that this structure was excited by the input transient current plotted in Fig. 5. It
results the graphs of H-field components displayed in Fig. 13.

Fig. 13. Maps of transient H-field components detected at t = 2 ns: (a) H
x
, (b) H
y
and (c) H
z

computed from the commercial tool CST


17
Similar to the previous section, these H-field components were mapped in the horizontal
plane placed at the height z
0
= 6 mm above the bottom surface of the considered filter and
delimited by -38 mm < x < 38 mm and -24 mm < y < 24 mm with space-step x = y = 2 mm.
By using the frequency-dependent data convoluted with the excitation test signal, we will
present next that we can regenerate these transient H-field maps.
4.2 Analysis of the results obtained from the transient field computation method
proposed

Fig. 14. Maps of frequency-dependent H-field components magnitude: (a) H
x
, (b) H
y
and (c)
H
z
computed from HFSS at f = 1 GHz


18
After HFSS-simulations carried out in the frequency range starting from f
min
= 0.05 GHz to
f
max
= 2.5 GHz with frequency-step f = 0.05 GHz, the maps of H-field component
magnitudes displayed in Fig. 14 are recorded. These field components are mapped in the
same horizontal plane as in the previous subsection by taking the height z
0
= 6 mm. Among
the series of the field maps obtained, we show here the data simulated at the frequency f = 1
GHz. According to the flow work depicted in Fig. 2, the frequency-dependent data H
x
(f),
H
y
(f) and H
z
(f) are employed for the determination of the time-dependent data H
x
(t), H
y
(t)
and H
z
(t) regarding the transient input current plotted earlier in Fig. 5. So that by
application of the computation algorithm under investigation, the H-field maps calculated
from Matlab are respectively, depicted in Figs. 15 at the instant time t = 2 ns.

Fig. 15. Maps of H-field components obtained from the proposed method and regarding the
simulated frequency-dependent data from HFSS: (a) H
x
, (b) H
y
and (c) H
z


19
Despite the slight difference of H
x
-maps, we observe that the maps are perfectly well-
correlated to those introduced in Fig. 13 of subsection 4.1. For the further smart illustration
of the results correlation, comparisons of H-field profiles calculated with the method
proposed (grey curves) and those from CST (black curves) for x = -5 mm are plotted Fig. 16.

Fig. 16. Comparisons between Oy-profiles of the H-field components computed with CST
software and those obtained from the proposed method
The imperfection of the results presented here are due to the numerical errors mainly caused
by the solver and the meshing inaccuracies. Note that one evaluates relative errors of about
10 % for |H
x
|, |H
y
| and |H
z
|. Despite the apparent difference between the results from the
method under investigation and the commercial EM-tool CST-computations, once again,
very good correlations between the profiles of the H-field components are realized by
considering the data recorded in the vertical cut-plane equated by x = -5 mm as explained in
Fig. 16.
In nutshell, the computation results exposed in this paper reveal the effectiveness and the
operability of the method developed for the case of elementary magnetic dipoles and also by
considering the NF EM-radiation of realistic use case electronic devices.
5. Concluding remarks
A computation method of transient NF EM-field radiated by electronic devices excited by a
complex wave or ultra-short duration transient signal is stated in this chapter. In addition to
the evanescent wave integration, the originality of the NF calculation method developed lies
on the consideration of the radiation deeming the UWB structures which is literally from DC
to microwave frequency ranges. It is based on the convolution of the frequency-dependent


20
EM-wave data with a transient excitation pulse current. A methodological analysis was
made by taking into account a complex waveform of the transient pulse signal exciting the
radiation source structure considered. It was explained how the frequency-bandwidth of the
frequency-dependent baseband EM-field must be chosen according to the excitation current
considered.
In order to demonstrate the relevance of the method investigated, it was first, implemented
into Matlab program and then, applied to the determination of the H-field radiated by a
microwave circuit in UWB. As consequence, the feasibility of the method was verified with
two types of structures. First, with the semi-analytical calculation implemented in Matlab by
considering the frequency- and time-dependent expressions of the magnetic NF radiated by
a set of magnetic dipoles, an excellent agreement with the results from the calculation
method developed were found. Then, further more practical analysis was performed with
the determination of transient H-NF from the frequency-dependent data computed with a
standard commercial 3-D EM-tool. For this second test, the H-NF emitted by a low-pass
planar microstrip filter was treated. For both cases, the excitation current injected to the
structures was assumed as an ultra-short transient pulse having half-bandwidth lower than
5 ns which presents a baseband frequency spectrum with bandwidth of about 2.5 GHz from
DC. With the examples of complex structures tested, very good agreement between the
transient H-field component maps and profiles was realized from the method proposed and
those directly calculated from the well-known standard tools and from classical
mathematical EM-formulae.
It is interesting to point up that the NF computation method introduced in this chapter is
advantageous in terms of:
1. Simplicity of the EM-field maps determination for any waveform of transient excitation
even with ultra-short duration which is very hard to simulate with most of commercial
simulation tools. The method developed can be used for the determination of the NF
maps in time-domain which is practically very difficult to measure in the realistic
contexts.
2. It is flexible for various types of excitation signals which can be expressed analytically
and also from the realistic use case of disturbing signal generally met in EMC area
(Wiles 2003, Liu, K. 2011, Hubing 2011).
3. It can be adapted also to different forms of electrical and electronic structures for low-
and high-frequency applications. Globally speaking, it offers a possibility to work in
UWB from DC to unlimited upper frequency limit.
4. One can achieve significant EM-field measurement in very short time-duration with
base band measured data in wide bandwidth.
However, its main drawback is the limitation in term of time step which depends on the
frequency range of the initial frequency-data considered and also the necessity of powerful
computer for the achievement of high accurate results.
In the next step of this work, we plane to extend this method to transpose in time-domain
the modelling of EM-radiation with the optimized association of elementary dipoles (Vives-
Gilabert et al. 2009, Fernndez-Lpez et al. 2009). Then, we are hopeful that the method
developed in this chapter is very helpful for EMC/EMI investigations of modern
electrical/electronic systems as the case of hybrid vehicle embedded circuits (Vye 2011,


21
Trscher 2011, Kopp 2011) where the transient NF effects are susceptible to disturb the
system functioning.
6. Appendix
This appendix contains two parts of theoretical parts concerning the transient NF radiated
by the elementary magnetic dipoles (Baum 1971 & 1976, Singaraju & Baum 1976, Ravelo et
al. 2011a & 2011b, Lui Y. et al. 2011a & 2011b) and the bi-exponential signal processing.
6.1 Appendix A: Analytical study of the magnetic dipole radiation in time-domain
By definition, the magnetic dipole moment of the elementary circular loop shown earlier in
Fig. 3 (see section 3) is written as:
( , ) ( ) ( )
z
M
p r t p t r u =
G G
, (A-1)
with r is the distance between the dipole centre and the point M(r,,) as shown in Fig. 3. By
analogy with the definition of the time-variant vector established by Hertz in 1892 (Hertz
1892), the H-field components in the spherical coordinate system:
( , , , ) ( , , , ) ( , , , )
r
r
H H r t u H r t u H r t u

= + +
JJG G G G
, (A-2)
are expressed as:

2
( ) ( ) cos( ) 1
( , , , )
2
r
p p
H r t
r v t r

= +

, (A-3)

2
2 2 2
( ) ( ) ( ) sin( ) 1
( , , , )
4
p p p r
H r t
r v t r v t

= + +

, (A-4)
( , , , ) 0 H r t
= , (A-5)
where v is the wave-velocity, and is the time delayed variable which is defined as
/ t r v = . One underlines that the magnetic dipole is also an Hertzian dipole so that
( , ) / 0 i r t r = . In the frequency domain, the spherical coordinate of the H-field component
formulas radiated by the magnetic dipole pictured in Fig. 3 which is supposed flowed by an
harmonic current with amplitude I
M
denoted:

2
( )
j ft j t
M M
I f I e I e

= = , (A-6)
are written as (Balanis 2005):
( )
2
3
( , , , ) 1 cos( )
2
jkr M
r
I a
H r f jkr e
r

= + , (A-8)

2
2 2
3
sin( )
( , , , ) (1 )
4
jkr M
I a
H r f jkr k r e
r

= + , (A-9)


22
( , , , ) 0 H r f
= , (A-10)
where j is the complex number 1 and the real 2 / k f v = expresses the wave number at
the considered frequency f. Then, through the classical relationship between the spherical
and cartesian coordinate systems, one can determine easily the expressions of the
components H
x
, H
y
and H
z
.
6.2 Appendix B: Spectrum analysis of bi-exponential signal
A bi-exponential form signal with parameters
1
and
2
is analytically expressed as:

1 2
/ /
( ) ( )
t t
M
i t I e e

= . (B-1)
The analytical Fourier transform expression of this current is written as:

1 2
1 2
( ) ( )
1 1
M
I I
j j

=
+ +
, (B-2)
with is the angular frequency. This yields the signal frequency spectrum formulation
expressed as:

1 2
2 2 2 2
1 2
( )
(1 )(1 )
M
I I

=
+ +
. (B-3)
To achieve at least 95-% of excitation signal spectrum energy, the frequency-data should be
recorded in baseband frequency range with angular frequency bandwidth equal to:

2 2 2 2 2 2 2
1 2 1 2 1 2
95%
1 2
( ) 1600
2

+
= . (B-4)
7. Acknowledgment
Acknowledgement is made to EU (European Union) and Upper Normandy region for the
support of these researches. These works have been implemented within the frame of the
Time Domain Electromagnetic Characterisation and Simulation for EMC (TECS) project
No 4081 which is part-funded by the Upper Normandy Region and the ERDF via the
Franco-British Interreg IVA programme.
8. References
Adada, M. (2007). High-Frequency Simulation Technologies-Focused on Specific High-
Frequency Design Applications. Microwave Engineering Europe, (Jun. 2007), pp. 16-
17
Agilent EEsof EDA. (2008). Overview: Electromagnetic Design System (EMDS), (Sep. 2008)
[Online]. Available from:
http://www.agilent.com/find/eesof-emds


23
Ansoft corporation. (2006). Simulation Software: High-performance Signal and Power Integrity,
Internal Report
ANSYS, (2009). Unparalleled Advancements in Signal- and Power-Integrity, Electromagnetic
Compatibility Testing, (Jun. 16 2009) [Online]. Available from:
http://investors.ansys.com/
Archambeault, B.; Brench, C. & Connor, S. (2010). Review of Printed-Circuit-Board Level
EMI/EMC Issues and Tools. IEEE Trans. EMC, (May 2010), Vol. 52, No.2, pp. 455-
461, ISSN 0018-9375
Balanis, C. A. (2005). Antenna Theory: Analysis and Design, in Wiley, (3rd Ed.), 207208, New
York, USA, ISBN: 978-0-471-66782-7
Barriere, P.-A.; Laurin, J.-J. & Goussard, Y. (2009). Mapping of Equivalent Currents on High-
Speed Digital Printed Circuit Boards Based on Near-Field Measurements. IEEE
Trans. EMC, (Aug. 2009), Vol.51, No.3, pp. 649 - 658, ISSN 0018-9375
Baudry, D.; Arcambal, C.; Louis, A.; Mazari, B. & Eudeline, P. (2007). Applications of the
Near-Field Techniques in EMC Investigations. IEEE Trans. EMC, (Aug. 2007),
Vol.49, No.3, pp. 485-493, ISSN 0018-9375
Baum, C. E. (1971). Some Characteristics of Electric and Magnetic Dipole Antennas for Radiating
Transient Pulses, Sensor and Simulation Note 405, 23 Jan. 71
Baum, C. E. (1976). Emerging Technology for Transient and Broad-Band Analysis and
Synthesis of Antennas and Scaterrers, Interaction Note 300, Proceedings of IEEE,
(Nov. 1976), pp. 1598-1616
Braun, S.; Glten, E.; Frech, A. & Russer, P. (2009). Automated Measurement of Intermittent
Signals using a Time-Domain EMI Measurement System, Proceedings of IEEE Int.
Symp. EMC, pp. 232-235, ISBN 978-1-4244-4266-9, Austin, Texas (USA), Aug. 17-21
2009
Chew, W. C. & Kong, J. A. (1981). Electromagnetic field of a dipole on a two-layer earth.
Geophysics, (Mar. 1981), Vol. 46, No. 3, pp. 309-315
Cicchetti, R. (1991). Transient Analysis of Radiated Field from Electric Dipoles and
Microstrip Lines. IEEE Trans. Ant. Prop., (Jul. 1991), Vol.39, No.7, pp. 910-918, ISSN
0018-926X
Edwards, R. S.; Marvin, A. C. & Porter, S. J. (2010). Uncertainty Analyses in the Finite-
Difference Time-Domain Method. IEEE Trans. EMC, (Feb. 2010), Vol.52, No.1, pp.
155-163, ISSN 0018-9375
Fernndez-Lpez, P.; Arcambal, C.; Baudry, D.; Verdeyme, S. & Mazari, B. (2009). Radiation
Modeling and Electromagnetic Simulation of an Active Circuit, Proceedings of EMC
Compo 09, Toulouse, France, Nov. 17-19 2009
Hertz, H. R. (1892). Untersuchungen ueber die Ausbreitung der Elektrischen Kraft (in
German). Johann Ambrosius Barth, Leipzig, Germany, ISBN-10: 1142281167/ISBN-
13: 978-1142281168
Hubing, T. (2011). Ensuring the Electromagnetic Compatibility of Safety Critical
Automotive Systems. Invited Plenary Speaker at the 2011 APEMC, Jeju, South-
Korea, May 2011
Jauregui, R.; Pous, M.; Fernndez, M. & Silva, F. (2010). Transient Perturbation Analysis in
Digital Radio, Proceedings of IEEE Int. Symp. EMC, pp. 263-268, ISBN 978-1-4244-
6307-7, Fort Lauderdale, Florida (USA), Jul. 25-30 2010


24
Jauregui, R.; Riu, P. I. & Silva, F. (2010). Transient FDTD Simulation Validation, Proceedings
of IEEE Int. Symp. EMC, pp. 257-262, ISBN 978-1-4244-6305-3, Fort Lauderdale,
Florida (USA), Jul. 25-30 2010
Jun-Hong, W.; Lang, J. & Shui-Sheng, J. (1997). Optimization of the Dipole Shapes for
Maximum Peak Values of the Radiating Pulse, Proceedings of IEEE Tran. Ant. Prop.
Society Int. Symp., Vol.1, pp. 526-529, Montreal, Que., Canada, 13-18 Jul 1997, ISBN
0-7803-4178-3
Kopp, M. (2011). Automotive EMI/EMC Simulation. Microwave Journal, (Jul. 2011), Vol.54,
No.7, pp. 24-32
Lakhtakiaa, A.; Varadana, V. K. & Varadana, V. V. (1987). Time-Harmonic and Time-
Dependent Radiation by Bifractal Dipole Arrays. Int. J. Electronics, (Dec. 1987),
Vol.63, No.6, pp. 819-824, DOI:10.1080/00207218708939187
Liu, K. (2011). An Update on Automotive EMC Testing. Microwave Journal, (Jul. 2011),
Vol.54, No.7, pp. 40-46
Liu, L.; Cui, X. & Qi, L.. (2009). Simulation of Electromagnetic Transients of the Bus Bar in
Substation by the Time-Domain Finite-Element Method. IEEE Trans. EMC, (Nov.
2009), Vol.51, No.4, pp. 1017-1025, ISSN 0018-9375
Liu, Y.; Ravelo, B.; Jastrzebski, A. K. & Ben Hadj Slama, J. (2011). Calculation of the
Time Domain z-Component of the EM-Near-Field from the x- and y-
Components, Accepted for communication in EuMC 2011, Manchester, UK, Oct. 9-
14 2011
Liu, Y.; Ravelo, B.; Jastrzebski, A. K. & Ben Hadj Slama, J. (2011). Computational
Method of Extraction of the 3D E-Field from the 2D H-Near-Field using PWS
Transform, Accepted for communication in EMC Europe 2011, York, UK, Sep. 26-30
2011
North East Systems Associates (NESA), (2010). RJ45 Interconnect Signal Integrity, (2010 CST
Computer Simulation Technology AG.) [Online]. Available from:
http://www.cst.com/Content/Applications/Article/Article.aspx?id=243
Ordas, T.; Lisart, M.; Sicard, E.; Maurine, P. & Torres, L. (2009). Near-Field Mapping
System to Scan in Time Domain the Magnetic Emissions of Integrated Circuits,
Proceedings of PATMOS 08: Int. Workshop on Power and Timing Modeling
Optimization and Simulation, Ver. 1-11, Lisbon, Portugal, Sep. 10-12 2008, ISBN
978-3-540-95947-2
Rammal, R.; Lalande, M.; Martinod, E.; Feix, N.; Jouvet, M.; Andrieu, J. & Jecko, B. (2009).
Far Field Reconstruction from Transient Near-Field Measurement Using
Cylindrical Modal Development, Int. J. Ant. Prop., Hindawi, Vol. 2009, Article ID
798473, 7 pages, doi:10.1155/2009/798473
Ravelo, B. (2010). E-Field Extraction from H-Near-Field in Time-Domain by using PWS
Method. PIER B Journal, Vol.25, pp. 171-189, doi:10.2528
Ravelo, B.; Liu, Y.; Louis, A. & Jastrzebski, A. K. (2011). Study of high-frequency
electromagnetic transients radiated by electric dipoles in near-field. IET Microw.
Antennas Propag., (Apr. 2011), Vol. 5, No, 6, pp 692 - 698, ISSN 1751-8725
Ravelo, B.; Liu, Y. & Slama, J. B. H. (2011). Time-Domain Planar Near-Field/Near-Field
Transforms with PWS Method. Eur. Phys. J. Appl. Phys. (EPJAP), (Feb. 2011), Vol.53,
No.1, 30701-pp. 1-8, doi: 10.1051/epjap/2011100447


25
Rioult, J.; Seetharamdoo, D. & Heddebaut, M. (2009). Novel Electromagnetic Field
Measuring Instrument with Real-Time Visualization, Proceedings of IEEE Int.
Symp. EMC, pp. 133-138, ISBN 978-1-4244-4266-9, Austin, Texas (USA), Aug. 17-
21 2009
Schantz, H. G. (2001). Electromagnetic Energy around Hertzian Dipoles. IEEE Tran. Ant.
Prop. Magazine, (Apr. 2001), Vol.43, No.2, pp. 50-62, ISBN 9780470688625
Selin, V. I. (2001). Asymptotics of the Electromagnetic Field Generated by a Point Source in a
Layered Medium. Computational Mathematics and Mathematical Physics, Vol.41, No.6,
pp. 915-939, ISSN 0965-5425
Shi, J.; Cracraft, M. A.; Zhang J. & DuBroff, R. E. (1989). Using Near-Field Scanning to
Predict Radiated Fields, Proceedings of IEEE Ant. Prop. Int. Symp., Vol.3, pp. 1477-
1480, San Jose, CA (USA)
Singaraju, B. K. & Baum, C. E. (1976). A Simple Technique for Obtaining the Near Fields of
Electric Dipole Antennas from Their Far Fields, Sensor and Simulation Note 213, Mar.
76
Smagin, S. I. & Mazalov, V. N. (2005). Calculation of the Electromagnetic Fields of Dipole
Sources in Layered Media. Doklady Physics, (Apr. 2005), Vol.50, No.4, pp. 178-183,
DOI:10.1134/1.1922556
Song, J. & Chen, K.-M. (1993). Propagation of EM Pulses Excited by an Electric Dipole in a
Conducting Medium. IEEE Tran. Ant. Prop., (Oct. 1993), Vol.41, No.10, pp. 1414-
1421, ISSN 0018-926X
Song, Z.; Donglin, S.; Duval, F.; Louis, A. & Fei, D. (2010). A Novel Electromagnetic
Radiated Emission Source Identification Methodology. Proceedings of Asia-Pacific
Symposium on EMC, Pekin (China), Apr. 12-16 2010, ISBN 978-1-4244-5621-5
Sten, J. C.-E. & Hujanen, A. (2006). Aspects on the Phase Delay and Phase Velocity in the
Electromagnetic Near-Field. PIER Journal, Vol. 56, pp. 67-80, doi:10.2528
Trscher, M. (2011). 3D EMC/EMI Simulation of Automotive Multimedia Systems.
Microwave Journal, (Jul. 2011), Vol.54, No.7, pp. 34-38
Vives-Gilabert, Y.; Arcambal, C.; Louis, A.; Daran, F.; Eudeline, P. & Mazari, B. (2007).
Modeling Magnetic Radiations of Electronic Circuits using Near-Field Scanning
Method. IEEE Tran. EMC, (May 2007), Vol.49, No.2, pp. 391-400, ISSN 0018-
9375
Vives-Gilabert, Y.; Arcambal, C.; Louis, A.; Eudeline, P. & Mazari, B. (2009). Modeling
Magnetic Emissions Combining Image Processing and an Optimization Algorithm.
IEEE Tran. EMC, (Nov. 2009), Vol.51, No.4, pp. 909-918, ISSN 0018-9375
Vye, D. (2011). EMI by the Dashboard Light. Microwave Journal, (Jul. 2011), Vol.54, No.7, pp.
20-23
Wiles, M. (2003). An Overview of Automotive EMC Testing Facilities, Proceedings of
Automotive EMC Conf. 2003, Milton Keynes, UK, Nov. 6 2003
Winter, W. & Herbrig, M. (2009). Time Domain Measurement in Automotive Applications,
Proceedings of IEEE Int. Symp. EMC, pp. 109-115, ISBN 9781424442669, Austin, Texas
(USA), Aug. 17-21 2009
Xie, L. & Lei, Y. (2009). Transient Response of a Multiconductor Transmission Line With
Nonlinear Terminations Excited by an Electric Dipole. IEEE Trans. EMC, (Aug.
2009), Vol.51, No.3, pp. 805-810, ISSN 0018-9375


26
Yang, T.; Bayram, Y. & Volakis, J. L. (2010). Hybrid Analysis of Electromagnetic Interference
Effects on Microwave Active Circuits Within Cavity Enclosures. IEEE Trans. EMC,
(Aug. 2010), Vol.52, No.3, pp. 745-748, ISSN 0018-9375
0
Impulse-Regime Analysis of Novel
Optically-Inspired Phenomena at Microwaves
J. Sebastian Gomez-Diaz
1
, Alejandro Alvarez-Melcon
1
,
Shulabh Gupta
2
and Christophe Caloz
2
1
Universidad Politcnica de Cartagena
2
cole Polytechnique de Montral
1
Spain
2
Canada
1. Introduction
The ever increasing of needs for high data-rate wireless system is currently producing a shift
from narrow-band radio towards ultra-wideband (UWB) radio operation [see Ghavami et al.
(2007)]. Novel microwave tools, concepts, phenomena and direct applications must be
developed to meet this demand. While the past decades have been focused on the "magnitude
engineering" and lter design [see Pozar (2005)], there is a renewed interest in the "dispersion
engineering". In the dispersion engineering approach, the phase is engineered to met various
specications within a given frequency range, so as to process signals in real time.
In this context, the development of electromagnetic metamaterials over the last years [see
Caloz & Itoh (2006) or Marques et al. (2008)], with their intrinsically dispersive nature and
subsequent impulse-regime properties, may provide novel and original solutions (see Fig. 1).
Metamaterials can easily be synthesized in planar technology under the form of composite
right/left-handed (CRLH) transmission lines (TLs), using non-resonant [see Caloz & Itoh
(2006)] or resonant [see Duran-Sindreu et al. (2009)] approaches. These structures have
provided novel and exciting applications, such as multi-band components, diplexers,
couplers, phase-shifters, power-dividers or antennas with enhanced features, to mention
just a few [see Caloz (2009) or Eleftheriades (2009) for a recent review]. However, CRLH
structures have mostly been analyzed in the harmonic regime to date, and therefore only a
few impulse-regime components and systems have been proposed so far. An example of
these applications is the tunable pulse delay line presented in in Abielmona et al. (2007).
In this chapter, we present recent advances based on Fourier transformation techniques to
model dispersive UWB phenomena and far-eld radiation from complex CRLH structures.
Section 2 rst employs inverse Fourier transforms to study pulse propagation along this
type of medium. Then, a Fourier transform approach is applied to the current which ows
along the CRLH line, accurately retrieving the time-domain far-eld radiation of the structure
[which behaves as a leaky-wave antenna, (LWA)]. The main advantages of the proposed
techniques are the easy treatment of complex CRLHstructures, a deep insight into the physics
of the phenomena, and an accurate and a fast computation, which avoids the time-consuming
analysis required by completely numerical simulations.
2
2 Will-be-set-by-IN-TECH
Fig. 1. Illustration of the dispersion engineering concept using metamaterial and CRLH
structures. Reprinted with permission from Abielmona et al. (2008). Copyright 2008, URSI.
Section 3 applies the previously derived theory to study the impulse-regime phenomenology
of CRLH structures, and subsequently demonstrates several optically-inspired phenomena
and applications at microwaves in both the guided and the radiative regime. This study is
divided into two main groups, according to the guided-wave or radiative-wave natures of the
proposed phenomena and applications.
In the guided-regime, the dispersive properties of pulse propagation along a CRLH line and
the temporal Talbot effect, which allows the period rate multiplication of an input pulse
train [see Gmez-Daz et al. (2009b)], will be described and analyzed.
In the radiative regime, the spectral-spatial decomposition property of CRLH LWAs will be
exploited to characterize both, in time and in frequency, unknown fast-transient input
signals. For this purpose, two different systems (namely, a real time spectrogram analyzer
(RTSA) [see Gupta, Abielmona & Caloz (2009)], and a frequency resolved electrical gating
(FREG) [see Gupta, Gmez-Daz & Caloz (2009)]) will be carefully studied and simulated.
In all cases, the proposed theory and phenomena will be validated by using simulation
results from full-wave commercial softwares and measurements, from fabricated prototypes.
Therefore, the usefulness of the Fourier transform approach will be fully demonstrated as
an essential mathematical tool for the fast and accurate modeling of very complex UWB
structures and phenomena.
2. Impulse-regime analysis of CRLH structures using a Fourier transform
approach
Propagation of electromagnetic short pulses in complex media, see Felsen (1969), has been a
eld of great interest for a long time. Most practical developments for pulse propagation in
dispersive media have been carried out for optical systems, including optical bers, couplers,
switches and soliton devices [see Saleh & Teich (2007)]. At microwaves, pulse propagation
has been far less studied, and the temporal analysis of dispersive metamaterial structures is
usually performedwith time-domain full-wave methods. However, these accurate techniques
require a high computational cost, due to the meshing of the whole geometry under study.
28 Fourier Transform Applications
Impulse-Regime Analysis of Novel Optically-Inspired Phenomena at Microwaves 3
In this section, a general time-domain Greens function approach is presented for the
analysis of pulse propagation in electrically thin CRLH TLs. This method is based on the
transient analysis of 1D transmission lines [see Paul (2007)] combined for the rst time with
CRLH TL concepts introduced in Caloz & Itoh (2006). With this equivalent transmission
line simplication of the geometry, the Greens functions are available in closed-form, and
directly correspond to the voltages and currents along the transmission line. The main
advantages of this approach are the unconditional stability and fast computation, due to the
continuous treatment of time, and the insight into the physical phenomena provided by the
Greens functions. Subsequently, the method is extended to analyze impulse-regime CRLH
leaky-wave antennas. The approach is based on the use of the time-domain current which
ows along the structure to compute the far-eld radiation of the antenna. This technique is
especially appropriate to characterize complex radiated-wave UWB phenomena and devices,
as it will be demonstrated in Section 3.
2.1 Composite right/left-handed structures
The introduction of metamaterials [see Caloz & Itoh (2006) or Marques et al. (2008)] in the last
decade has paved the road to the development of new devices and applications based on the
novel fundamental features and phenomena associated to this type of media. Among the most
useful metamaterials, one can nd the CRLH transmission lines [see Caloz & Itoh (2006)].
This type of transmission lines, which are inherently nonresonant and low-loss, can be easily
implemented in planar technology (such as microstrip or coplanar waveguide, for instance)
and provides a practical realization of electromagnetic metamaterials. As any metamaterial,
CRLH TLs are generally periodic structures formed by the repetition of unit-cells (an example
of this type of cells is shown in Fig. 2) whose size, p, must fulll the condition p
g
(where
g
is the guided wavelength) ir order to emulate an effectively homogeneous material.
A powerful method to analyze these metamaterial lines is the TL approach, presented in
Caloz & Itoh (2006), which employs an ABCD-matrix technique of periodically arranged
unit-cells to model the articial transmission line (see Fig. 3) and to determine its wave
propagation characteristics (such as propagation constant or Bloch impedances, see Fig. 4).
There are many examples of interesting and groundbreaking applications of planar TL
metamaterials at microwaves, such as for instance multi-band components, lters and
diplexers, couplers, power-dividers, phase-shifters, lenses, or backre to endre leaky-wave
antennas. A review of these and much more applications and devices can be found in
metamaterials textbooks, such as in Eleftheriades & Balmain (2005), in Caloz & Itoh (2006),
or in Marques et al. (2008). Note that all previously mentioned and most of metamaterials
applications operate in the harmonic regime up to now, and they have been designed for
narrow-band components and systems (even though some of them may support a multi-band
operation).
2.2 Impulse regime analysis of CRLH transmission lines
Many media, ranging from traditional purely right-handed materials to recent CRLH
metamaterials [see Caloz & Itoh (2006)], can be advantageously analyzed by the transmission
line theory described in Pozar (2005). So far, this theory has been applied mostly in the
harmonic regime, where Greens functions for both the voltage and the current along the
line are available. However, it may also be employed in the time-domain, where the
Greens function approach provides an efcient tool to analyze impulse regime signals along
29 Impulse-Regime Analysis of Novel Optically-Inspired Phenomena at Microwaves
(a) (b)
Fig. 2. Equivalent unit cell circuit model of a lossless CRLH transmission line. (a)
Asymmetric conguration. (b) Symmetric conguration.
Fig. 3. Equivalence between N cascaded unit cells and a transmission line of length ,
characterized by an equivalent complex propagation constant
0
and Bloch impedance Z
0
.
200 150 100 50 0 50 100 150 200
1
1.5
2
2.5
3
3.5
4
4.5
5
x 10
10
Propagation constant [()]
BF
Light Line
RH Range LH Range
EF
cR
cL
(a)
0 25 50 70
1
1.5
2
2.5
3
3.5
4
4.5
5
x 10
10
Bloch Impedance []
cL
cR
(b)
Fig. 4. Dispersion diagram (a) and frequency-dependent Bloch impedance (b) related to a
balanced CRLH unit cell. The size of the unit cell is p = 1 cm and its circuital parameters are
C
R
= C
L
= 1.0 pF and L
L
= L
R
= 2.5 nH. The frequencies
cL
and
cR
denote the bandpass
frequency region of the line.
transmission lines. In this case, the point source model accurately characterizes a pulse
generator, and the computed quantities are the voltages and currents along the line as a
function of time.
Consider an electric source

J(r, t) placed in an arbitrary homogeneous and dispersive
medium. The wave equation in this case reads [see Collin (1991)]
E(r, t) +

2
t
2
E(r, t) =

t
J(r, t). (1)

The spatial-temporal dyadic Greens function G(r,r
; t, t
) in this equation for a specic

medium and a specic source is obtained as the response to a unitary point source
J(r
, t
) =
(r
; t
). As described in Barton (1989), once this Greens function is known, the electric eld
may be computed using
E(r, t) =
_ _

G(r, r
g

; t, t
J( r
g

, t
)d r
g

dt
, (2)
where the spatial-temporal Greens function may be expressed in terms of its inverse Fourier
transform
G(r, r
g

; t, t
) =
1
2
_
+
G(r, r
g

; )e
j(tt
)
d. (3)
Inserting Eq. (3) into Eq. (2), yields
E(r, t) =
1
2
_ _ _

G(r, r
g

; )
J( r
g

, t
)e
j(tt
)
dr
g
dt
d. (4)
This expression provides the eld radiated by an arbitrary source (in space and time) in an
arbitrary dispersive homogenous medium. Since the spatial-temporal distribution of the
source is generally known, only the Greens function needs to be computed to provide the
eld solution. An analogous formulation may naturally be obtained for the magnetic eld.
In case of electrically thin 1D transmission lines (placed along the z direction, as shown in
Fig. 5), the generator source may be reducedto a point source, greatly reducing the complexity
of the problem. Consider a point source placed at the position r
g
, and with a temporal
dependence
J(r
g
, t
) =(r
g
)I
g
(t
) = (r r
g
)I
g
(t
) e
z
. (5)
In this case Eq. (4) is reduced to
E(r, t) =
1
2
_ _

G(r, r
g

; ) I
g
(t
) e
z
e
j(tt
)
dt
d. (6)
At this point, we use the Fourier transform of the temporal source [
I
g
() = F{I
g
(t
)}, where
the operator F denotes a Fourier transform, see Pipes & Harvill (1971)]. Note that the input
pulse is usually modulated at a frequency
0
, which is included in the

I
g
() notation as a
e
j
0
t
term. Besides, a transmission-line Greens function is used to obtain the voltage (V) or
the current (I) along the 1D line, which may be expressed as
X(z, t) =
1
2
_

G
X
(z, z
g
; )

I
g
() e
jt
d, (7)
where z
g
is the source position (r
g
= z
g
e
z
), z is the observation point and X(z, t) denotes the
voltage or current along the line (in the z direction), as a function of the Greens functions
employed [

G
V
(z, z
g
; ) or

G
I
(z, z
g
; ), related to the voltage or current, respectively]. It
should be noted that the space dependence has been absorbed in the Greens function term,
while the temporal information is described by Fourier and inverse-Fourier transforms.
Let us consider a simple matched transmission line, as shown in Fig. 5(a) [Z
g
= Z
L
=
Z
0
(), ]. In this simple case, the transmission line Greens functions for the voltages and
(a)
(b)
(c)
Fig. 5. Dispersive articial transmission line excited by a point source generator. (a) Uniform
case. The line, composed of N unit cells, is dened by its characteristic impedance [Z
0
()],
complex propagation constant [()] and length (). (b) Non-uniform case. The line is
composed of N uniform transmission line sections. Each k
th
section has its own length (
k
),
characteristic impedance [Z
0
k
()] and propagation constant [
k
()]. (c) Thvenin
equivalent circuit for the k
th
uniform transmission line section. Reprinted with permission
from Gmez-Daz et al. (2010). Copyright 2010, IET.
currents may be expressed as
G
V
(r,r
g
; ) = e
()R
, (8)
G
I
(r,r
g
; ) =
e
()R
Z
0
()
, (9)
respectively, where () is the complex propagation constant (or dispersion relation), Z
0
()
is the characteristic impedance, and R = |z z
g
| is the distance between the observation point
z along the line and the source point z
g
(generator).
Using these expressions, the voltage or the current along the line can easily be found
with Eq. (7). Note that this equation applies to any type of transmission line, including
metamaterial CRLH lines, provided that the propagation constant [()] is known.
Consider now the more general case of a nonuniform transmission line medium composed of
N uniform transmission line sections (or unit-cells), as shown in Fig. 5(b). The sections may
be different from each other and may be of different type. Therefore, reections occur due to
the transition between two consecutive cells, and different propagation conditions appear at
each cell. The Greens function along the k
th
uniform transmission line section (z [
k
, 0],
possibly innitesimal), including generator and load mismatches, reads
G
k
(z, z
= ; ) = A()
_
e
k
()z
+
l,k
()e
k
()z
_
, (10)
where [see Pozar (2005)]
A() =
V
Th,k
()Z
in,k
()
Z
in,k
() + Z
Th,k
e
k
()
k
1
l,k
()
Th,k
()e
2
k
()
k
, (11)
and
Th,k
() =
Z
Th,k
() Z
0
k
()
Z
Th,k
() + Z
0
k
()
, (12)
l,k
() =
Z
in,k+1
() Z
0
k
()
Z
in,k+1
() + Z
0
k
()
, (13)
which was obtained by equating the Greens function evaluated at the input of the k
th
section,
to the Thvenins voltage evaluated at the same section V
Th,k
[see Fig. 5(c)]. By repeatedly
applying Eq. (10) from the load to the generator or vice-versa, the voltage as a function of
time may be computed at any point along the nonuniform transmission medium.
Finally, note that the use of periodic input signals is important to produce some periodic
phenomena or devices, such as the Talbot effect introduced in Azaa & Muriel (2001) or the
UWB resonator presented in Gmez-Daz et al. (2009a), among many others. For this purpose,
the theory previously introduced can easily be extended to consider this type of input signals.
In this case, the input signal may be represented in the time domain by
I
p
g
(t) =
k=+
k=
I
g
(t kT
0
), (14)
where T
0
is the period rate. The voltage or current along the transmission line may then be
computed by inserting Eq. (14) in Eq. (7). Note that for many input pulses, such as modulated
Gaussian pulses, the interchange between the integral and summation operations (related to
Eq. (7) and Eq. (14), respectively), allowed by the assumed linearity of the system, will further
contribute to reduce the computational cost required by this technique.
Fig. 6. Illustration of a CRLH LWA. The antenna can be congured to radiate at backwards
[ <
T
and () < 0], forwards [ >
T
and () > 0] or broadside [ =
T
and
() = 0].
2.3 Impulse regime analysis of CRLH leaky-wave antennas
A CRLH transmission line supports a fast-wave mode [see Oliner & Jackson (2007)] which
penetrates inside the fast-wave region. Therefore a CRLH structure behaves as a leaky-wave
antenna when it is excited by a source with a frequency within a range of (
BF
< <
EF
),
where
BF
and
EF
are the fast-wave region limits [see Fig. 4(a) and Caloz & Itoh (2006)].
Since a CRLH line behaves as a LWA, the direction of the radiated main beam follows the
LWA scanning law, which is given by [see Oliner & Jackson (2007)]
sin()
()
k
0
. (15)
In the above equation, is the radiation angle (measured from the perpendicular direction
over the CRLH structure), () is the phase constant, and k
0
is the free-space wavenumber.
Fig. 6 presents an illustration related to the operation principle of a CRLH LWA. As can be
seen in the gure, and following Eq. (15), the antenna is able to radiate at backwards [when
<
T
and () < 0], forwards [ >
T
and () > 0] and broadside [ =
T
and
() = 0]. Therefore, this type of structure is able to provide a full-space scanning, from
backre ( = 90
) to endre ( = +90
), including the broadside ( = 0
) direction. The
use of CRLH transmission lines as leaky-wave antennas has led to the development of many
radiated-wave applications, most of themin the harmonic regime [see Caloz & Itoh (2006) and
Eleftheriades & Balmain (2005)].
This subsection proposes an impulse-regime analysis of leaky-wave structures. Even though
the proposed study is valid for all types of LWAs, we will focus on CRLH LWA structures
because they are broadband in nature and they are able to radiate from backre to endre,
including the broadside direction [see Caloz & Itoh (2006)]. As previously pointed out, a
CRLH leaky-wave antenna follows the beam-scanning law of Eq. (15). Therefore, each input
frequency, which lies inside the fast-wave region, is radiated towards a different direction into
space. This situation is explicitly depicted in Fig. 7(a). As can be seen in the gure, there is a
unique correspondence between each input frequency [
x
, with (
BF

x

EF
)] and its
associated radiation angle (
x
). Therefore, each frequency is mapped into a different angle in
space.
According to Eq. (15), if a CRLH LWA is excited by a modulated input pulse, as shown
in Fig. 7(b), each spectral component of the signal is radiated towards a different direction
(a) (b)
Fig. 7. Impulse-regime behavior of CRLH LWAs. (a) Frequency-space relationship of a CRLH
LWA. The dispersion curve is graphically related to its corresponding beam scanning law. (b)
Spectral decomposition of a pulse obtained by the frequency-space mapping property of a
CRLH leaky-wave antenna. Reprinted with permission from Gupta, Abielmona & Caloz
(2009). Copyright 2009, IEEE.
in space at any particular instant. Hence, the CRLH LWA performs an instantaneous
spectral-to-spatial decomposition of the input pulse. This decomposition allows to discriminate
the various spectral components present in the input signal. In this sense, there is clear
parallelism between LWAs (which usually operates at microwaves) and diffraction gratings
[see Saleh & Teich (2007)], which usually operate at the optics regime and radiate each spatial
input frequency towards a different direction angle in space. The main advantage of CRLH
LWAover diffraction gratings is its simple point feeding systemas compared to the diffraction
gratings, which require plane-wave illumination.
Let us consider a single 1D (or electrically thin) LWA, located along the z axis, as shown in
Fig. 8. It is important to clarify the notation employed to describe the situation under analysis.
First, the line is dened by a characteristic impedance [Z
0
()], a complex propagation
constant [() = () + j()], and a total length, . Second, the generator which excites
the transmission line is placed at the positionr
g
, withr
g
= z
g
e
z
. Third, any point along the
line is denoted as r
, withr
= z
e
z
(note that z
start
z
z
end
, see Fig. 8). Besides, note
that shall study the far-eld radiation of the transmission line towards an observation point P
(denoted asr, withr = x e
x
+y e
y
+z e
z
). Therefore, any observation point P must be located in
the far-eld region of the transmission line [see Balanis (2005)], fullling the far-eld radiation
condition, which is given by
R
ant
=
x
2
+ y
2
+
_
z

2
_
2
>
2
2
0
. (16)
In this last equation, R
ant
is the distance between the observation point P (placed at r) and
the transmission line, which can approximately be considered as a point source (placed at the
center of the line) from a far-eld point of view.
Fig. 8. Sketch of a single 1D CRLH LWA. The electrically thin antenna is considered as a
linear wire from a far-eld point of view. It is placed along the z-axis, it has a length of
= z
end
z
start
, and it is fed by a point generator, placed atr = z
g
e
z
.
Let us also assume that the CRLHLWA is excited by a modulated input pulse. In this case, the
time-domain theory developed in Section 2.2 for the analysis of impulse-regime CRLHTL can
be employed to obtain the current which ows along the structure, which may be expressed
as
I(z
, t) =
_

G
I
(z
, z
g
, )
I
g
()e
jt
d. (17)
In this last equation,

G
I
(z
, z
g
, ) represents the transmission line Greens function, related to
the current, and

I
g
() denotes the Fourier transform of the temporal input pulse.
The electric eld radiated by the transmission line under study, in the far-eld region, is
approximately given by
E(r, ) j
A(r, ), (18)
where

A is the magnetic vector potential. It is important to keep in mind that this potential is
related to the physical current which is owing on the transmission line [see Balanis (2005)].
In the harmonic case, this potential may be expressed as
A(r, ) =

0
4
_
z
end
z
start
I(z
, )
e
jk
0
R
R
dz
e
z
, (19)
where R is the distance between the pair of source-observation points.
However, in the impulse-regime case the CRLH LWA is excited by a temporal pulse. This
time-domain behavior of the current induced on the structures propagates into the magnetic
vector potential, which now reads
A(r, t) =

0
4
_

EF
BF
_
z
end
z
start
I(z
, )
e
jk
0
R
R
ddz
e
z
. (20)
It is important to remark that the frequency integration limits have been modied with respect
to Eq. (7). Specically, the frequency limits directly corresponds to the fast-wave region range
(
BF

EF
). This is because LWAs are only able to radiate spectral components which
lies inside this region. Outside of the fast-wave region, this type of structures behaves as a
transmission lines as described in Caloz & Itoh (2006).
Finally, the time-domain electric eld radiated by a CRLH LWA, under a far-eld assumption,
can easily be recovered using
E(r, t) =
j
0
4
_

EF
BF
_
z
end
z
start
I(z
, )
e
jk
0
R
R
ddz
e
z
. (21)
This expression provides the time-domain electric eld radiation from a CRLH LWA excited
by a modulated input pulse, at any observation point placed in the far-eld region. The main
features of this closed-formformulation are:
All CRLH LWA radiation features at far-eld are taken into account by using a time-domain
current along the structure. This current is closely related to the complex propagation
constant of the structure.
Physical insight into the antenna radiation properties. The Greens function and the
current owing along the structure, related to the CRLH structure, completely dene the
antenna behavior. These parameters are obtained in closed-form.
Extremely fast computation, because most expressions are simple and well-behaved
integrals (and some of them are analytical for specic input pulses).
Capability to deal with any type of input pulse, providing a continuous temporal output
with unconditional stability.
The proposed approach can characterize complex radiated-wave UWB phenomena and
devices. Specically, this formulation will be employed in the next section to model several
systems, such as a real-time spectrum analyzer (RTSA) or a frequency-resolved electrical
gating system (FREG). It will be shown that the proposed formulation is able to completely
characterize these systems with high efciency and accuracy .
3. Optically-inspired phenomena at microwaves
3.1 Introduction
This section explores the impulse-regime phenomenology of CRLH structures and the
subsequent theoretical and practical demonstration of several novel optically-inspired
phenomena and applications at microwaves, in both, the guidedand the radiative regime. The
time-domain Greens function approach introduced in Section 2 has opened the door to a very
fast, but still accurate, analysis of these novel microwave phenomena and applications, most
of them transported from optics, exploiting either the group velocity or the group velocity
dispersion parameters of CRLH TL. The study can be divided into two main groups, related
to the guided-wave or radiative-wave natures of the proposed phenomena and applications.
The analogy between these phenomena at microwave and their corresponding counterpart at optics [see
Saleh & Teich (2007)] is deduced from the dispersive properties of CRLH structures. Specically, in
the guided mode there is a clear parallelism between the dispersive behavior of a CRLH line
and an optical component (such as an optical ber), which is inherently dispersive. Therefore,
optical phenomena [such as the well-known temporal Talbot effect] can be reproduced at
microwaves. In the radiative mode, the beam scanning law of the CRLH LWA is analog
to a diffraction grating, where different spectral components are radiated (or diffracted) at
different angles causing spatial dispersion. This may be exploited to develop spectrogram
analyzers, capable to fully characterize both, in time and frequency, any unknown input
signal.
3.2 Guided-regime
This subsection presents a study on the phenomenology of pulse propagation along CRLH
transmission lines. As previously stated, the CRLH TL represents a general transmission
medium, which is highly dispersive, especially in the left-handed frequency range. Next,
the temporal Talbot effect [see Azaa & Muriel (1999)] is introduced, theoretically described
and numerically conrmed for the case of CRLH media.
3.2.1 Pulse propagation along CRLH structures
The main goal of this subsection is to experimentally validate the theory presented in
Section 2. For this purpose, the propagation of a modulated pulse along a CRLHTL is studied.
The tunability of the pulse delay as a function of the modulation frequency is demonstrated
by measuring the temporal delay from different modulated pulses, leading to a tunable delay
system[see Abielmona et al. (2007)]. Then, the dispersive features of the CRLH TL are further
demonstrated monitoring the effects of pulse propagation along a matched and mismatched
line, in a cell-by-cell fashion.
First, we consider a CRLH transmission line composed of 30 unit cells and with the circuit
parameters C
R
= 1.8 pF, C
L
= 0.9 pF, L
R
= 3.8 nH and L
L
= 1.9 nH (transition frequency
f
0
= 2.55 GHz), excited by a modulated Gaussian pulse [ = 3.0 ns, following the notation
introduced in Saleh & Teich (2007)]. Fig. 9 shows the time-delayed waveforms obtained by
the proposed theory for different carrier frequencies, and by experiments using a real-time
oscilloscope (Agilent InniiumDS0871204B). Excellent agreement is observed between theory
and experiments. The group delay response is an important parameter in dispersive systems
for analog signal processing applications. This parameter may be computed either in the time
domain by determining the time differences between the maxima of the input and output
pulses, or in the frequency domain by taking the derivative of the unwrapped phase of
the transmission scattering parameter S
21
. Fig. 10 shows the group delay along the same
CRLH line and for the same pulse as in Fig. 9, computed by the discussed theory using
the rst approach, and validated by experiment using both approaches. Again, very good
agreement is observed between theory and experiment. The small discrepancies between
the two measured results may be explained by the tolerance in the localization of the pulse
maxima. As it may be seen fromthese results and was originally presented in Abielmona et al.
(2007), the CRLH transmission line acts as an impulse tunable delay system.
To better visualize the dispersion of a pulse along a CRLH dispersive medium, consider now
a CRLHtransmission line twice as long as before (60 unit cells), but with the same parameters.
To completely validate the discussedtheory, this time the line is excited by a modulated square
pulse [ f
0
= 2.05 GHz, T = 2.2 ns, following the notation of Saleh & Teich (2007)]. An ABCD
matrix approach, described in Caloz & Itoh (2006), is employed to compute the propagation
constant of the line, taking into account the nite number of unit cells in the experiment. The
position-time trajectory of the pulse is presented in Fig. 11. Fig. 11(a) shows the computed
Fig. 9. Time-delayed Gaussian waveforms at the input/output of a CRLH transmission line
for different carrier frequencies, obtained with the method proposed in Section 2.2.
Measurement results are also shown for validation. The manufactured CRLH transmission
line is shown in the inset. Reprinted with permission from Gmez-Daz et al. (2009b).
Copyright 2009, IEEE.
1.8 2 2.2 2.4 2.6 2.8 3 3.2 3.4
4
5
6
7
8
9
10
11
Frequency [GHz]
D
e
l
a
y

[
n
s
e
c
]

Proposed theory, from pulse maxima locations in time
Measured, from pulse maxima locations on oscilloscope
From derivation of measured unwrapped phase of S
21
()
4,68
9.12
Fig. 10. Time delay versus modulation frequency, using the time difference between the
maxima of the input and output pulses along the same CRLH line as in Fig. 9. Measured data
using both, the same procedure as before and unwrapping the phase of S
21
(), are also
shown for validation. The delays obtained in Fig. 9 for f
c
= 3.2 GHz and f
c
= 1.9 GHz,
which are 4.68 and 9.12 ns, correspond to the two highlighted points. Reprinted with
permission from Gmez-Daz et al. (2009b). Copyright 2009, IEEE.
results, from which two observations may be made: i) at the end of the line, the temporal
width of the pulse has increased by a factor of 5 (at 50% of the magnitude), ii) the edges of the
square envelope have been rounded off by the band-pass ltering response of the CRLH line.
The small ripples near the end of the structure are explained by the Gibbs effect on the input
pulse due to the nite computational interval and resolution. Fig. 11(b) shows the measured
result using a high-impedance probe connected to the oscilloscope. The abrupt decrease of
the voltage magnitude after the 30
th
cell is due to the fact that the 60-cell experimental line is
(a) (b)
Fig. 11. Propagation of a modulated square pulse ( f
0
= 2.05 GHz, T = 2.2 ns) along a
matched CRLH transmission line. The line includes 60 unit cells of length p = 2.0 cm and the
circuital parameters are C
R
= 1.8 pF, C
L
= 0.9 pf, L
R
= 3.8 nH and L
L
= 1.9 nH. (a)
Simulation. (b) Measurement. Reprinted with permission from Gmez-Daz et al. (2009b).
(a) (b)
Fig. 12. Propagation of a modulated square pulse ( f
0
= 2.05 GHz, T = 2.2 ns) along an
open-ended CRLH transmission line. The line is identical to that of Fig. 11 except that it
includes only 30 unit cells. (a) Simulation. (b) Measurement. Reprinted with permission from
Gmez-Daz et al. (2009b). Copyright 2009, IEEE.
in fact constituted of two cascaded 30-cell lines with a small loss in the interconnection. The
agreement with theory is reasonable, considering the tolerances of the measurement setup.
Finally, let us investigate the effects of the reection of a pulse at a discontinuity of a CRLH
transmission line. For this purpose, the line is open-ended and it is excited by the same
modulated square pulse as in the previous case. The propagation and the reection of the
pulse as a function of space and time is presented in Fig. 12. Again, good agreement is
observed between the theory [Fig. 12(a)] and the experiment [Fig. 12(b)]. It is worth noticing
that an interesting interference pattern, between propagating and reected pulses, occurs near
the discontinuity.
3.2.2 Temporal Talbot effect
The Talbot effect is a periodic constructive interference pattern produced by a dispersive
transmission medium with second-order dispersion for a periodic input signal. It was rst
reported by H. F. Talbot in 1836 for the case of a source with periodic spatial variation
[see Talbot (1836)]. The temporal counterpart of this effect [see Azaa & Muriel (1999) or
Azaa & Muriel (2001)] occurs when a time-periodic signal is propagating along the same
kind of medium. An input pulse train with temporal period T
0
and pulse width T is
replicated at the position nz
T
(n N), where z
T
is called the Talbot distance (or self-imaging
distance). Also, an increased repetition rate of m pulses per T
0
is obtained at the fractional
distances z
f
= (s/m)z
T
(where s, m N), provided that (s/m) is an irreducible fraction,
and under the condition that: m < T
0
/T [see Azaa & Muriel (2001)]. The temporal
Talbot effect has been employed mainly in the optical regime, for applications such as the
generation of signals with ultrahigh repetition rate (THz) fromslower ranges (GHz), as shown
in Azaa & Muriel (2001), or pulse compression, as presented in detail in Berger et al. (2004).
At microwaves, this phenomenon is more difcult to reproduce because it is difcult to obtain
a second-order dispersive mediumat this frequency region. However, the recent introduction
of CRLHTLs [see Caloz & Itoh (2006)], which provides a dispersive broadband behavior, may
lead to novel scenarios where the temporal Talbot effect and their subsequent applications
can be reproduced. This section provides a theoretical demonstration of the temporal Talbot
effect in CRLH media, including a full-wave validation of the phenomena at microwaves.
Furthermore, the conditions for the phenomena existence in CRLH media are stated, in
connection with potential device applications.
Intuitively, the temporal Talbot effect occurs when a periodic pulse signal is transmitted
through a second order dispersion medium where the neighboring pulses interfere as a result
of dispersion so as to produce new temporal components. For the analysis, let us consider
a single modulated pulse, denoted by (t) =
0
(t)e
j
0
t
, where
0
(t) is a slowly varying
envelope. The periodic signal is then represented in the time domain, at the position z = 0, as
A(z = 0, t) =
n=+
n=
(t nT
0
), (22)
where T
0
is the repetition rate of the signal. In the spectral domain, the periodic signal
becomes discrete, and it may be expressed as
A(z = 0, ) =
r
n=+
n=
( = n
r
)( n
r
), (23)
where
r
= 2/T
0
is the spectral repetition frequency.
On the other hand, the transfer function of a lossless CRLH TL is given by
H(z, ) = e
j()z
, (24)
where () is the CRLH TL propagation constant. This propagation constant, around the
CRLH TL transition frequency, may be approximated as
() =

, (25)
where
R
= p/
L
R
C
R
and
L
= 1/
_
p
2
L
L
C
L
, as described in Caloz & Itoh (2006). In the
right handed side of Eq. (25), the rst term provides a simple time delay (or linear frequency
phase), whereas the second order term is responsible for the line dispersion. Eq. (25) can be
further expanded, employing Taylor series, around a modulation frequency (
0
) as
() =
0
+
1
(
0
) +
1
2
2
(
0
)
2
+O(
3
), (26)
where the termO(
3
) is related to the order of the error committed in the approximation, and
the term
n
is dened as
n
=
_
n
()
n
_
=
0
. (27)
Note that Eq. (26) is only valid for the case of narrowband pulses (centered at the frequency
0
and with bandwidth ).
Employing Eq. (26), the transfer function of the CRLH TL
_

H(
) =

H( =
0
+
) , where
a change of variable has been introduced from to
] takes the form
H(z,
) = exp
_
j
_
0
+
1
+
1
2
2
_
z
_
. (28)
In order to derive the Talbot distance, only the third expansion term of the exponential is
considered. This is because the rst two terms do not provide any information related to the
Talbot distance [which is only due to second order dispersion, see Azaa & Muriel (2001)].
Specically, the rst term is related to the modulation frequency of the pulse and does not
carry any information about the envelope, and the second term represent the group delay (or
retarded frame) of the signal.
Using Eq. (23) and Eq. (28), the spectrum of the signal at the distance z can be written as
A(z,
) =
r
n=+
n=
exp
_
j
2
z
2
_

(n
r
)(
n
r
) = (29)
r
n=+
n=
exp
_
j
2
z
2
_
2n
T
0
_
2
_

(n
r
)(
n
r
).
Note that the appearance of
squared term in this equation is due to the quadratic phase

factor in the spectral response of the CRLH TL dispersive medium.
Eq. (29) reveals that the Talbot effect [ i.e.,

A
z
(z, ) =

A(z = 0, ) ] occurs under the condition
2
z
2
_
2n
T
0
_
2
= p, (30)
where p N. The case of the rst integer Talbot distance (p = 1) is given by
z
T
=
T
2
0
2|
2
|
=
T
2
0
3
0
4
2
L
, (31)
where the identity
2
= 2
2
L
/
3
0
[see Eq. (27)] has been employed.
As demonstrated in Azaa & Muriel (2001), the Talbot distance can also be obtained at
fractionary distances given by z
f
= (s/m)zT, where s and m are irreducible integers. At
this fractionary distance, the periodic input signal is also self-imaged but with an increase
repetition rate by a factor of m. This phenomenon corresponds to the fractionary Talbot effect
[see Azaa & Muriel (2001)]. It is important to note that in the case of a CRLH TL, the Talbot
distance can be tuned externally by modifying the parameters T
0
and
0
, without changing
the intrinsic parameters of the line.
(a)
210 220 230 240 250
0
0.2
0.4
0.6
0.8
1
1.2
N
o
r
m
a
l
i
z
e
d

v
o
l
t
a
g
e
Time [nsec]

Input
Output from TDGF
Output from ADS
(b)
210 220 230 240 250
0
0.2
0.4
0.6
0.8
1
1.2
N
o
r
m
a
l
i
z
e
d

v
o
l
t
a
g
e
Time [nsec]

Input
Output from TDGF
Output from ADS
(c)
210 220 230 240 250
0
0.2
0.4
0.6
0.8
1
1.2
N
o
r
m
a
l
i
z
e
d

v
o
l
t
a
g
e
Time [nsec]

Input
Output from TDGF
Output from ADS
(d)
Fig. 13. Talbot repetition rate multiplication effect. a) CRLH TL with length corresponding to
the basic Talbot distance z
T
. b) Reconstruction of the original pulse train at the Talbot
distance z
T
. c) Repetition rate doubling at the distance z
T
/2. d) Repetition rate tripling at the
distance z
T
/3. Results obtained from the proposed approach, and validated using the
commercial software ADS. Reprinted with permission from Gmez-Daz et al. (2009b).
After theoretically deriving the Talbot distance related to CRLH TLs, the phenomena will
be veried using the numerical technique proposed in Section 2.2 and commercial full-wave
simulations. Consider a balanced lossless CRLH transmission line with circuital parameters
C
R
= C
L
= 1.0 pF and L
R
= L
L
= 2.5 nH. A modulated train of Gaussian pulses, with
temporal width of = 0.75 ns and period rate of T
0
= 8 ns is used to excite the line shown
in Fig. 13(a). Fig. 13(b) presents the input pulse train and the output pulse train at the Talbot
distance z
T
. The reconstruction of the initial train of pulses is conrmed, although a small
disagreement, due to higher order terms (greater than 2) of the CRLH dispersion relation,
is observed between two consecutive pulses. Fig. 13(c) and Fig. 13(d) show the input and
output voltages at the fractional Talbot distances of z
T
/2 and z
T
/3, respectively. The effect
of pulse multiplication is thus clearly conrmed, while the distortion is smaller because less
higher-order dispersion effects occur over a shorter distance of propagation.
The practical implementation of Talbot devices based on CRLH transmission lines depends
on the technology employed. For instance, with the CRLH parameters used in Fig. 13, a
microstrip implementation with a typical unit cell size of 1 cm would lead to a Talbot distance
z
T
of around 17 meters. This is unpractical, specially due to the losses. However, using
multilayer technology [see Horii et al. (2005)], this distance may be dramatically reduced to
sizes in the order of several centimeters, while even lower sizes apply for pulse rate repetition
multiplication. This decrease in length also decreases the total amount of losses, allowing
the Talbot effect to be applied in practical situations, such as the generation of signals with
ultrahigh repetition rate or pulse compression [see Berger et al. (2004)].
3.3 Radiative-regime
The full-space scanning capabilities of CRLH LWAs are exploited in this Section to
achieve the spectral-spatial decomposition of an input broadband signal. This property
is then applied to the development of a real-time spectrum analyzer [RTSA, see
Gupta, Abielmona & Caloz (2009)] and a frequency resolved electrical gating system
[FREG, see Gupta, Gmez-Daz & Caloz (2009)], which are UWB applications able to fully
characterize, in both time and frequency, an unknown signal. These systems are fully modeled
by the technique presented in Section 2.3, which constitutes an ideal tool to provide not only
a fast numerical system characterization, but a deep physical insight into the electromagnetic
properties of the devices.
3.3.1 Real time spectrogram analyzer (RTSA)
A real time spectrogram analyzer (RTSA) is a device able to provide a joint time-frequency
representation of an unknown input signal. This device is specially useful in modern UWB
systems, as described in Ghavami et al. (2007), where ultra-fast transient signals are involved.
The common output of a RTSAsystemis an spectrogram [see Cohen (1989)], which is a 2-Dplot
of a signal where the energy distribution is related to an image in a time-frequency plane. The
joint time-frequency representation provides information related to the temporal evolution of
each spectral component, and the exact amplitude and location in time of each frequency. It is
important to mention that spectrograms suffer from the fundamental "uncertainty principle"
limitation, which states
tf
1
2
, (32)
where f is the bandwidth of the gated signal and t is the gate duration. This implies an
inherent trade off between time and frequency resolution for all spectrograms.
There are two main approaches to obtain the spectrogram of an unknown input test signal
at microwave frequencies. The rst and more usual technique is based on the use of digital
components, which performs the short-time Fourier transform (STFT) [see Oppenheim (1996)]
of the unknown input signal. However, the use of this approach at microwaves has the
important drawback of requiring fast processors and large memories, which limits its use
to input signals of just a few hundreds megahertz and temporal resolutions of only a few
microseconds. The second option, purely analog, is based on the use of a bank of lters, as
described in Amin & Feng (1995). The main disadvantage of this approach is that it requires a
very large number of channels and extremely narrow-band lters, which is specially difcult
to achieve at high frequencies. Consequently, this option is very complex and expensive.
In this context, a novel RTSA based on the spectral-spatial decomposition of CRLH LWAs was
proposed in Gupta, Abielmona & Caloz (2009). In this approach, the CRLH LWA provides an
analog implementation of the STFT, in a similar way as it has been implemented in the optical
regime. A schematic of the CRLH LWA RTSA system is shown in Fig. 14. The description of
the system behavior is as follows. First, the spectral-spatial decomposition of the CRLH LWA
Fig. 14. Analog real-time spectrogramanalyzer showing the CRLH LWA, the antenna probes,
the envelope detectors, the A/D converters, the DSP block, and the display with the
spectrogram. Reprinted with permission from Gupta, Abielmona & Caloz (2009). Copyright
2009, IEEE.
is employedto discriminate the frequency components of the input test signal (see Section 2.3).
Second, a set of probes (antenna receivers) monitor the time variation of each frequency
component. Finally, a postprocessing step performs the analog/digital (A/D) conversion,
the data processing and the display of the spectrogram.
The analog CRLH LWA RTSA system provides several advantages and benets as compared
with other RTSAs at microwave frequencies. First, this approach is completely analog and
real time. Therefore, there is not requirement of large memories and fast processors, just a
light postprocessing stage. Second, the same RTSA system can use different CRLH LWAs, in
order to be exible and to cover several frequency ranges. The use of new technologies to
fabricate CRLH LWAs allows the use of a wide variety of input signals, from microwave up
to potentially millimeter-wave frequencies. Third, the CRLH LWA RTSA system is inherently
broadband, and it can be designed to process the 100% of an input signal bandwidth.
On the other hand, the CRLH LWA RTSA systemhas also to deal with some drawbacks. First,
it requires a far-eld probe conguration, which makes the system relatively large. However,
the system can possibly be compacted employing near-eld to far-eld transformations.
Second, the physical length of the LWA represents a space-gating mechanism which controls
both, the temporal and the frequency resolution of the resulting spectrograms. This is clearly
shown in Fig. 15, where a slice of the signals spatial waveform is presented on and radiating
from the antenna. And third, the time and frequency resolution also depends on the number
of detectors, and their associated response time and sampling frequency. Therefore, there is a
fundamental tradeoff between time and frequency resolutions.
In order to characterize this system, full-wave commercial software may be employed.
However, the generation of the output results are extremely time consuming due to the
complexity of the system and to the temporal nature of the analysis. An interesting and
efcient alternative is to employ the theory developed for the modeling of impulse-regime
radiation (see Section 2.3), to fully characterize the CRLH LWA RTSA system. For the sake of
Fig. 15. Impact of the LWA size on the time-frequency resolution of the spectrograms
generated by the an analog CRLH LWA RTSA.
validation, let us consider a complete RTSAsystembased on a CRLHLWAwhich is composed
of 32 unit cells of length p = 1.0 cm, with circuital parameters of C
R
= C
L
= 1.0 pF and
L
R
= L
L
= 2.5 nH. In order to complete the analog RTSA system, a total of number of 181
observation probes are placed in a semi-circular conguration, following the conguration of
Fig. 14. The rst step to model the RTSA is to perform the calibration of the system. This is
necessary to compensate for the different power levels received at each probe [as described
in Gmez-Daz et al. (2010)], due to the directivity variation with frequency of this type of
antennas. For the calibration, a narrow-band signal is modulated to the different fast-wave
frequencies [following Eq. (15) (scanning law)] and subsequently radiated by the LWA. Then,
the maximum power received at each probe is stored, obtaining a normalization rule for this
particular system conguration. In our example, the calibration data is shown in Fig. 16(a).
After the RTSA system has been calibrated, it can be efciently used to obtain spectrograms
of an unknown input signal. For the method validation, the actual temporal and frequency
information of an input test signal, previously known, is employed.
Let us consider now that the CRLH LWA is fed by a signal composed of three modulated
Gaussian pulses. The rst pulse has a positive-chirp modulation (which means that the
modulation frequency is increasing with time), and the third pulse has a negative-chirp
modulation. The spectrogram obtained with the proposed method, after calibration, is
depicted in Fig. 16(b), including an additional graph showing the analytical temporal
representation of the signal. As can be observed in the gure, the spectrogram follows the
signal variations in time (the three pulses are clearly observable) and also, simultaneously,
in frequency. It is especially interesting to observe the transition between two consecutive
pulses, where frequencies corresponding to different pulses appear at the same instant.
The method proposed here is able to perform a quick (about 30 seconds) and accurate
modeling of the RTSA system, with deep insight into the CRLH LWA time-radiation
properties, and avoiding the extremely time-consuming analysis required in full-wave
simulations (usually between 8 12 hours). Therefore, it provides a fast tool to congure
an RTSA system and to determine a priory the range of input signals which can accurately be
characterized for a given CRLH LWA.
90 60 30 0 30 60 90
0
10
20
30
40
50
60
70
Angle []
M
a
x
i
m
u
m

e
l
e
c
t
r
i
c

f
i
e
l
d

[
V
/
m
]
(a) (b)
Fig. 16. Results obtained from an RTSA system based on a CRLH LWA composed of 32 unit
cells of length p = 1.0 cm, with circuital parameters of C
R
= C
L
= 1.0pF and
L
R
= L
L
= 2.5nH. (a) Maximum electric eld obtained at the different positions of the
probes, used for the calibration of the RTSA system. (b) Normalized spectrogram of a three
chirp-modulated Gaussian pulses signal, with chirp parameters C = [10, 0, 10], modulation
frequency f
0
= 3.19 GHz and temporal width = 1.0 ns, computed with the proposed
technique. The inset shows the analytical time response of the signal. Reprinted with
permission from Gmez-Daz et al. (2010). Copyright 2010, IET.
Finally, experimental results from a RTSA prototype are included for a complete system
validation and to conrm the accuracy of the proposed modeling technique. For this
purpose, a CRLH LWA fabricated in microstrip technology and composed of 14 -0.8 cm-
long unit cells, with circuital parameters C
R
= 1.29 pF, C
L
= 0.602 pF, L
R
= 3.0 nH and
L
L
= 1.4 nH, is employed. A photo of the fabricated antenna is depicted in Fig. 17(a),
whereas a comparison of the measured and simulated scattering parameters is shown in
Fig. 17(b). Furthermore, Fig. 17(c) presents a simulation-measurements comparison of the
antenna dispersion relationship. As can be observed, the CRLH LWA presents its transition
frequency at 3.745 GHz. The antenna fast-wave frequency region starts at about 3.1 GHz
(related to backre radiation), and it is extended until 4.7 GHz (related to endre radiation).
Note that several resonances occur within the fast-wave frequency region (close to endre),
degrading the antenna performance. These resonances are due to internal resonances of the
interdigital capacitors employed in the antenna prototype, and they are not considered in the
circuital model of the LWA.
In order to measure the spectrogramof the test signals, a single receiver was used and rotated
around the circular far-eld trajectory of the system (specically, between = 80
and =
80
with increments of 5
). The RTSAsystemis calibrated employing a linear frequency ramp.

After calibration, the CRLH LWA is excited by a modulated Gaussian pulse, with FWHM
of 3.5 ns. Fig. 18 presents a comparison between simulations (obtained by the proposed
time-domain Greens function approach) and measurements, as a function of the pulse
modulation frequency. It can be observed that a very good agreement is achieved in all cases.
First, the pulse modulation frequency is set to 3.3 GHz, which corresponds to backwards
radiation. This is clearly shown in the spectrograms of Fig. 18(a) and Fig. 18(b). Then the
(a)
1 2 3 4 5
80
70
60
50
40
30
20
10
0
Frequency [GHz]
S
c
a
t
t
e
r
i
n
g

P
a
r
a
m
e
t
e
r
s

[
d
B
]

S11 Measured
S21 Measured
S11 Circuital
S21 Circuital
Fastwave region
Transition Frequency
(b)
400 300 200 100 0 100 200
1
1.5
2
2.5
3
3.5
4
4.5
5
[rad/m]
F
r
e
q
u
e
n
c
y

[
G
H
z
]

Circuital Model
Measurement
FreeSpace
(c)
Fig. 17. 1D CRLH LW antenna composed by 14 -0.8 cm- long cells, with circuital parameters
C
R
= 1.29 pF, C
L
= 0.602 pF, L
R
= 3.0 nH and L
L
= 1.4 nH. a) Photo of a microstrip CRLH
LWA prototype. b) Scattering parameters. c) Dispersion relation. Reprinted with permission
from Gmez-Daz et al. (2009c). Copyright 2009, American Institute of Physics.
modulation frequency is set to the CRLH LWA transition frequency, which corresponds to
broadside radiation. As expected, the obtained spectrograms [see Fig. 18(c) and Fig. 18(d)]
conrm the change in frequency. Finally, pulse modulation frequency is set to 4.2 GHz, which
corresponds to a forward direction. Again the spectrograms show the changes in the pulse
frequency, experimentally verifying the RTSA system and conrming the usefulness of the
proposed theoretical approach to model this type of analog systems.
3.3.2 Frequency resolved electrical gating system (FREG)
Analog CRLHLWARTSAs generate the spectrogramof an unknown input signal in real-time,
using the spectral-spatial decomposition property of the leaky-wave antenna, with minimal
requirements on computational resources. This system provides important benets over any
digital RTSAs at microwave frequencies. However, as explained in the previous section, the
time and frequency resolution of the generated spectrograms directly depends on the physical
length of the CRLH LWA, which is xed in a given system. Therefore, a particular antenna
can only handle a limited range of input signals, which must full specic time and frequency
constrains. This imposes an important limitation to the CRLH LWA RTSA systems.
This section proposes a novel analog approach to obtain spectrograms, where the hardware
dependence is suppressed, at the cost of the requirement of periodicity of the input signals.
This approach is inspired froma similar systemknown in optics as frequency resolved optical
gating (FROG) [see Trebino (2002)], where a self-gating principle is applied to provide close
to ideal spectrograms for arbitrary test signals. Here, we propose a microwave counterpart of
(a) (b)
(c) (d)
(e) (f)
Fig. 18. Spectrograms obtained by the proposed RTSA model (gures on the left) and by
experiments (gures on the right), employing the CRLH LWA of Fig. 17. A modulated
Gaussian pulse with FWHM = 3.5 ns feeds the antenna. The pulse modulation frequency is
set to 3.3, 3.745 and 4.2 GHz, corresponding to backward [(a) and (b)], broadside [(c) and (d)]
and forward [(e) and (f)] radiation, respectively.
Fig. 19. Proposed frequency resolved electrical gating (FREG) system. Reprinted with
permission from Gupta, Gmez-Daz & Caloz (2009). Copyright 2009, EuMC.
the FROG system, which is termed frequency resolved electrical gating (FREG). This system
is very useful for the measurement and characterization of fast-varying non-stationary UWB
signals and ultrashort pulses.
In order to compute a spectrogramof a signal x(t), a temporal gating function g(t) is required
[see Cohen (1989)]. Using a self-gating approach, instead of using a separate time signal as the
gate function, the envelope of the testing signal itself is used as the gating function, i.e. g(t) =
|x(t)|. The spectrogram of a signal x(t) is then obtained as
S(, ) =
x(t)|x(t )|e
jt
dt
2
. (33)
The proposed FREG system, based on this self-gating principle and on the spectral-spatial
decomposition property of the CRLH LWA, is depicted in Fig. 19. The testing signal, whose
spectrogram is to be generated, is split into two channels. One of the channels is envelope
detected and passed through a tunable delay line. The two channels are then mixed together.
The mixer thus performs the self-gating process at a given time delay instant . This self-gated
signal is then injected into a CRLH LWA which spectrally resolves it in space. Once the
frequency components are separated in space, antennas circularly placed in the far-eld of
the LWA receive the different frequency components corresponding their angular position.
All the received signals are then digitized and summed, before being stored for spectrogram
display. This process is repeated for different values of the time delay so that the entire
test signal is scanned, according to Eq. (33), until the spectrogram is fully constructed. Since,
the beam scanning law of CRLH LWA is nonlinear in nature, a nal post-processing step is
required to linearize the spectrogram.
The proposedsystemexhibits signicant advantages over the analog RTSAsystemand purely
digital systems. Due to the self-gating process, neither the time nor the frequency resolutions
of the generated spectrogram depend on the physical length of the antenna. The time and
frequency resolutions are thus dependent only on the time signal itself and the hardware
Fig. 20. Simulated spectrograms. a) Down-chirped gaussian pulse (C
1
= 10, C
2
= 0,
f
0
= 4 GHz). b) Non-chirped super-gaussian pulse (C
1
= C
2
= 0, f
0
= 3 GHz). c)
Up-chirped gaussian pulse (C
1
= +10, C
2
= 0, f
0
= 4 GHz). d) Cubically chirped gaussian
pulse (C
1
= 0, C
2
= 0.25 10
28
). All pulse have a FWHM duration of 1 ns with a initial pulse
offset of t
0
= 6.5 ns, and are described using Eq. (34). Reprinted with permission from
Gupta, Gmez-Daz & Caloz (2009). Copyright 2009, EuMC.
dependence spectrogram is suppressed. The LWA simply plays a role of spectral decomposer
which, when longer (higher directivity), provides better separation of frequencies in space.
The choice of the gate duration is an important parameter to achieve an optimal
time-frequency resolution in the spectrogram. An optimal gate duration for pulses with
dominating phase variations is given by T
g
1/
_
2|
(t)|, where
(t) is the second

time derivative of phase [see Cohen (1989)]. This duration permits the resolution of the
fastest phase variations. For general pulse measurement, a gate duration as short as the
testing signal itself or slightly shorter is thus desirable. Since the FREG system is based on
self-gating, the gate duration is close to optimal and the corresponding spectrograms are ideal,
as demonstrated in Trebino (2002).
Moreover, the proposed system being analog in nature, neither require fast processors nor
huge memory buffers, which avoid placing a heavy computational burden on the system.
Furthermore, the system is frequency scalable and sufciently broadband to handle a wide
variety of UWB signals. As mentioned above, the length of the LWA controls the spectral
decomposition of the gated signal, which is improved as the physical length of the antenna
is increased. Finally, since it uses a multi-shot measurement procedure, where the testing
signal is gated several times with different time delays , the proposed FREG system requires
a periodic input signal. This is the main limitation of the FREG system.
In order to simulate the proposed FREG system, the time-domain Greens functions approach
presented in Section 2.3 is employed. The role of this theory is to model the transient CRLH
LWA behavior, which provides the spectral-spatial decomposition property and is a key
component of the proposed FREG system. Then, the other components of the system are
implemented as follows. The envelope of the testing signals are numerically obtained and
used as a gating function. The tunable time delay between the replica of the test signal and
the gate signal is applied, and the mixer, which performs the self-gating operation, is modeled
by a simple mathematical product. The resulting signal is then fed into the CRLH LWA. As a
nal stage, the temporal radiation computed at the probe locations are integrated as a function
of the gate delay (), in order to recompose the desired spectrogram.
Once the numerical model of the system is complete, spectrograms obtained from various
test signals are computed. For this purpose, consider a 16-cell CRLH LWA with the circuital
parameters C
L
= C
R
= 1 pF, L
L
= L
R
= 2.5 nH and a unit cell size of p = 2 cm, easily
implemented in metal-insulator-metal (MIM) technology [see Abielmona et al. (2007)]. The
various modulated testing pulses are gaussian and super gaussian-type signals which follows
the equation
v(t) = C
0
Re
_
exp
_
j2f
0
t
1
2
(1 + jC
1
)
_
t t
0
_
2m
+ jC
2
(t t
0
)
3
__
, (34)
where C
0
controls the pulse amplitude, t
0
is the time offset, is related to the pulse duration,
C
1
and C
2
control the linear and quadric chirp modulation, f
0
is the pulse modulation
frequency and m is an integer.
Fig. 20 shows FREG-generated spectrograms. Figs. 20(a) and (c) show the spectrograms of
a down-chirped and up-chirped gaussian pulses, respectively. A faithful representation of
a linear instantaneous frequency variation is obtained. The spectrogram of a modulated
un-chirped super-gaussian pulse is shown in Fig 20(b), where the occurrence of all the
frequency components of the signal at the same time instant are clearly seen. Finally, Fig. 20(d)
shows the spectrogram of a cubically chirped (down and up) gaussian pulse. The high
frequency components occurring at two different times, characteristic of cubically chirped
pulses, can be clearly identied. These few examples demonstrate the capability of the
proposed FREG system to analyze a wide variety of non-stationary signals.
It is important to point out that a full-wave simulation of the FREG system is
extremely time-consuming. Specically, the FREG system requires multiple analysis of the
impulse-regime response of a CRLH LWA, fed by different input signals. Since each of
these analysis lasts between 8 10 hours, the simulation of a complete FREG spectrogram
may easily lasts few days, which is completely prohibitive. On the other hand, the use
of the time-domain Greens functions approach reduces this time to a few (5 8) minutes.
Furthermore, the use of this numerical tool provides a deep insight into the physics of the
system, including an electromagnetic modeling of the antenna and a clear understanding of
each step of the proposed FREG system.
Finally, a comparison between the FREG and the RTSA systems is given in Fig. 21. The goal
of this analysis is to point out the advantages and disadvantages of each method. For the
comparison, the same input pulse feeds the RTSA and FREG systems, which are based on
identical CRLH LWAs. In the comparison, the number of unit cells N (with size p = 1.56 cm
and circuital parameters of C
L
= C
R
= 1 pF and L
L
= L
R
= 2.5 nH) of the antenna is
modied to perform several tests. The spectrogramresults are then given as a function N, i.e.
as a function of the total length of the antenna ( = N p). For the test, a modulated Gaussian
pulse is employed (with f
0
= 3.0 GHz and = 0.5 ns). In the gure, the results fromthe FREG
system are placed on the left, whereas the spectrograms computed by the RTSA system are
located on the right. First, we set in both systems an antenna with N = 5 unit cells, obtaining
the spectrograms shown in Fig. 21(a) and Fig. 21(b). This antenna is physically very short,
which turns out into a very low directivity. This leads to a very bad frequency resolution
(a) (b)
(c) (d)
(e) (f)
Fig. 21. Spectrograms obtained by the proposed FREG (gures on the left) and RTSA (gures
on the right) systems, based on identical CRLH LWAs for the different tests. The antennas are
composed of different numbers of N cells, with length p = 1.56 cm and circuital parameters
of C
L
= C
R
= 1 pF and L
L
= L
R
= 2.5 nH. A modulated Gaussian pulse feeds the systems
( f
0
= 3.0 GHz, = 0.5 ns). The resulting spectrograms are given for the case of N = 5 [(a)
and (b)], N = 20 [(c) and (d)] and N = 40 [(e) and (f)] unit cells.
in both spectrograms. On the other hand, this antenna provides an excellent time-gating
trade-off, because the energy is instantaneously radiated, almost without propagation along
the structure, leading to a large time resolution. Thereby, the use of a very short antenna
leads to generally erroneous spectrograms, due to the wide detection of frequencies which
are not part of the input pulse. Second, we modify the CRLH LWA antenna, including now
a total of N = 20 unit cells. This conguration provides a good frequency resolution in
both systems, while the temporal resolution is deteriorated in the RTSA system (due to the
use of a longer antenna). The resulting spectrograms are depicted on Fig. 21(c) (FREG) and
Fig. 21(d) (RTSA). As it can be observed, the FREG system provides a completely realistic
spectrogram, which faithfully reproduces the input signal in terms of frequency and time
(location and spreading). On the other hand, the spectrogram obtained by the RTSA has a
good frequency resolution, but has some problems dealing with the temporal duration of the
pulse. As previously commented, this problem is due to the propagation of the input pulse
as it is being radiated, as graphically illustrated in Fig. 15. And third, we simulate the FREG
and RTSA systems based on the same CRLH LWA, but composed now of N = 40 unit cells.
The results are shown in Fig. 21(e) (FREG) and Fig. 21(f) (RTSA). The spectrogram obtained
using the FREG system is quite similar to the previous FREG spectrogram(N = 20 unit cells),
keeping the temporal characteristics but improving the frequency resolution (because a longer
antenna provides higher directivity). All relevant features of the input modulated Gaussian
pulse, in terms of frequency and time, can easily be extracted fromthis spectrogram. However,
the RTSA system provides a completely wrong result. This is because of the excessive length
of the CRLH LWA, which completely destroy the temporal resolution of the system.
The above comparison demonstrates that the proposed FREG system presents important
advantages over the RTSA system, specially in terms on temporal resolution, being able to
characterize any unknown UWB input signal. Furthermore, this comparison has shown that
the RTSA system can only deal with signals whose frequency and temporal characteristics
are -at least overall- previously known. On the other hand, the main constrains of the FREG
system are the complex equipment required, the requirement of a periodic input signal, and
the fact that it is not a completely real-time system.
4. Conclusions
This chapter has introduced an impulse-regime analysis of metamaterial-type transmission
lines and antennas. Specically, a novel formulation, based on Fourier transformations,
has been proposed to describe pulse propagation along dispersive linear CRLH lines. The
proposedtheory is capable to model complex impulse-regime phenomena, such as dispersion,
in a simple, accurate and fast way. Then, the method has been extended to consider
CRLH leaky-wave antennas, allowing a fast and accurate analysis of the far-eld radiation
of these structures in time-domain. The proposed formulation has then been applied to
the development of novel phenomena and applications in the microwave domain, most of them
transported from optics. Instead of the usual magnitude engineering and lter design, a
dispersion or phase engineering has been applied. In this approach, the dispersive nature
and subsequent impulse-regime properties of CRLH structures have been exploited to obtain
novel phenomena/applications. Each phenomenon or application proposed has theoretically
been described, numerically veried, and in most of the cases, experimentally demonstrated.
The shift from narrow band systems (mostly used in the past) to ultra wide band systems,
required by current high date rate wireless communication systems, suggests that the
forthcoming decades will experience a major interest on this dispersive engineering approach,
providing new, novel and more exciting effects and devices at microwaves.
5. References
Abielmona, S., Gupta, S. &Caloz, C. (2007). Experimental demonstration and characterization
of a tunable CRLH delay line system for impulse/continuous wave, IEEE Microwave
and Wireless Components Letters 17(12): 864866.
Abielmona, S., Gupta, S., Nguyen, H. V. & Caloz, C. (2008). Dispersion engineered
impulse regime memataterial devices, Proc. XXIXth Assembly of Union Radio Science
International (URSI), Chicago, IL, USA.
Amin, G. & Feng, K. D. (1995). Short-time fourier transforms using cascade lter
structures, IEEE Transaction on Circuits and Systems II, Analog Digital Signal Processing
20(3): 631641408.
Azaa, J. & Muriel, M. A. (1999). Technique for multiplying the repetition rates of periodic
trains of pulses by means of a temporal self-imaging effect in chirped ber gratings,
Optics Lett. 24: 16721674.
Azaa, J. & Muriel, M. A. (2001). Temporal self-imaging effects: theory and application for
multiplying pulse repetition rates, IEEE J. Sel. Top. Quantum Electron 7: 728744.
Balanis, C. A. (2005). Antenna Theory: Analysis and Design. 3rd Edition., John Wiley and Sons.
Barton, G. (1989). Elements of Greens Functions and Propagation, Oxford Science Publications.
Berger, N. K., Levit, B. S., Bekker, A. & Fischer, B. (2004). Compression of periodic
optical pulses using temporal fractional talbot effect, IEEE Photonics Technology Letters
16(8): 18551857.
Caloz, C. (2009). Perspectives on EM metamaterials, Materials Today 12(3).
Caloz, C. & Itoh, T. (2006). Electromagnetic Metamaterials: Transmission Line Theory and
Microwave Applications., Wiley and IEEE Press.
Cohen, L. (1989). Time-frequency distributions-a review, Proceedings of the IEEE 77(2): 941981.
Collin, R. E. (1991). Field Theory of Guided Waves, IEEE Press, Piscataway, N.J.
Duran-Sindreu, M., Velez, A., Aznar, F., Bonache, J. & Martin, F. (2009). Application of open
split ring resonators and open complementary split ring resonators to the synthesis
of articial transmission lines and microwave passive components, IEEE Transactions
on Microwave Theory and Techniques 57(2): 33953403.
Eleftheriades, G. (2009). EM transmission-line metamaterials, Materials Today 12(3).
Eleftheriades, G. V. & Balmain, K. G. (eds) (2005). Negative-Refraction Metamaterials:
Fundamental Principles and Applications, Wiley & IEEE Press, Hoboken, NJ.
Felsen, L. B. (1969). Transients in dispersive media, part I: Theory, IEEE Transactions on
Antennas and Propagation 17: 191200.
Ghavami, M., Michael, L. B. & Kohno, R. (2007). UWB Signals and Systems in Communication
Engineering, J. Wiley & Sons.
Gmez-Daz, J. S., Gupta, S., lvarez-Melcn, A. & Caloz, C. (2009a). Impulse-regime
CRLH resonator for tunable pulse rate multiplication, Radio Science
44(doi:10.1029/2008RS003991): 19.
Gmez-Daz, J. S., Gupta, S., lvarez-Melcn, A. & Caloz, C. (2009b). Investigation
on the phenomenology of impulse-regime metamaterial transmission lines, IEEE
Transactions on Antennas and Propagation 57(12): 40104014.
Gmez-Daz, J. S., Gupta, S., lvarez-Melcn, A. & Caloz, C. (2009c). Tunable talbot imaging
distance using an array of beam-steered metamaterial leaky-wave antennas, Journal
of Applied Physics 106: 0849089.
Gmez-Daz, J. S., Gupta, S., lvarez-Melcn, A. & Caloz, C. (2010). Effcient time-domain
anlisis of highly-dispersive linear and non-linear metamaterial waveguide and
antenna structures, IET Microwaves, Antennas and Propagation 4(10): 16171625.
Gupta, S., Abielmona, S. & Caloz, C. (2009). Microwave analog real-time spectrum analyzer
(rtsa) based on the spatial-spectral decomposition property of leaky-wave structures,
IEEE Transactions on Microwave Theory and Techniques 57(12): 40104014.
Gupta, S., Gmez-Daz, J. S. & Caloz, C. (2009). Frequency resolved electrical gating
principle for UWB signal characterization using leaky-wave structures, 39th European
Microwave Conference, Rome, Italy.
Horii, Y., Caloz, C. & Itoh, T. (2005). Super-compact multilayered left-handed transmission
line and diplexer application, IEEE Transactions on Microwave Theory and Techniques
53(4).
Marques, R., Martn, F. & Sorolla, M. (eds) (2008). Metamaterials with Negative Parameters:
Theory, Design and Microwave Applications, Wiley, Hoboken, NJ.
Oliner, A. A. & Jackson, D. R. (2007). Leaky-wave antennas, in J. L. Volakis (ed.), Antenna
Engineering Handbook, 4 edn, McGraw-Hill, New York.
Oppenheim, A. V. (1996). Signals and Systems, Prentice Hall.
Paul, C. R. (2007). Analysis of Multiconductor Transmission Lines, 2
nd
edition edn, Wiley-IEEE
Press.
Pipes, L. A. & Harvill, L. R. (1971). Applied Mathematics for Engineers and Physicist, 3
rd
edn,
McGraw Hill.
Pozar, D. (2005). Microwave Engineering, 3
rd
edn, John Wiley and Sons.
Saleh, B. E. A. & Teich, M. C. (2007). Fundamentals of Photonics, 2nd edition edn,
Wiley-Interscience.
Talbot, H. F. (1836). Facts relating to optical science no. IV, Philos. Mag. 9: 401407.
Trebino, R. (ed.) (2002). Frequency-Resolved Optical Gating: The Measurement of Ultrashort Laser
Pulses, Springer.
3
Fourier Transform Application
in the Computation of
Lightning Electromagnetic Field
Vesna Javor
University of Nis / Faculty of Electronic Engineering
Serbia
1. Introduction
Atmospheric discharge is one of the most interesting and powerful natural phenomenon
hiding from men its undiscovered features and secrets for centuries. Lightning discharges
have been studied in many theoretical and experimental ways. However, application of
Fourier transform has been introduced in this research only in recent few decades. It proved
very useful in solving many lightning research problems.
There are two main groups of problems in lightning studies that involve using Fourier
transform. One deals with determining how the energy is distributed over a continuous
frequency spectrum for the quantity of interest. Channel-base currents, induced voltages
and currents, electric and magnetic field components, so as integrals and derivatives of the
same functions, distribute components over the entire frequency range in different ways.
These are non-periodic functions scattering their energy throughout the frequency spectra.
Another important group of problems to be solved by Fourier transform deals with the
calculation of lightning induced effects at different distances from the lightning discharge
and risk assessment for buildings, various structures, people and property in an external
impulse electromagnetic field. These calculations are based on experimental results for the
measured lightning electromagnetic field (LEMF) and channel-base currents, which are used
in different types of lightning stroke models including lossy ground effects. Lightning
discharge channel is usually modeled by a thin vertical antenna at a lossy ground of known
electrical parameters. Both ground and air are treated as linear, isotropic and homogeneous
half-spaces. Even for such a simple approximation of the lightning channel - calculations can
not be easily done in time domain, but transformation to frequency domain is used instead.
Once the calculations are done in frequency domain, way back to time domain is made by
Inverse Fourier transform applied to the obtained results.
In both groups of problems an impulse function which has the analytical derivative, integral
and integral transformations is very useful. New functions proposed by the author for
representing lightning currents are presented in this Chapter, Section 2. These can be also
used in other high voltage technique calculations. The main problem for a user of the
impulse functions already given in literature to approximate some quantity is the choice of
parameters so to obtain desired waveshape characteristics or values adequate to


58
experimentally measured. Parameters of the new channel-base current (NCBC) function
(Javor, 2008; Javor & Rancic, 2011) are calculated according to IEC 62305 standard lightning
currents (International Electrotechnical Commission, Technical Committee 81 [IEC, TC 81],
2006), and the procedure to choose function parameters is also explained in (Javor & Rancic,
2011). Functions to represent other typical lightning currents are proposed in (Javor, 2011b,
2011c), such as long stroke current (LSC), and two-rise function (TRF) as a multi-peaked
current. These functions can be used to obtain the desired peak value, rise-time, decaying
time to half of the peak value, current steepness, integral of the function (representing also
impulse charge), or integral of the square of the function (representing also specific energy),
etc. Fourier transform is obtained analytically and the results are presented in Section 3.
Application of the Fourier transform in LEMF computation is shown in Section 4. Based on
these results, some conclusions are given in Section 5.
2. NCBC function
Any mathematical function capable to approximate the impulse lightning channel-base
current in electromagnetic field calculations can be also used to represent far lightning
electromagnetic field as an external excitation inducing currents and voltages in objects
and structures inside such a field. Such functions are necessary in lightning stroke
modeling.
There is an overview of lightning return-stroke models in (Rakov & Uman, 1998, 2006)
where they are classified into four classes according to the type of governing equations. An
adequate return-stroke model would be the one that provides simultaneously an
approximation to the experimentally measured channel-base current, to the lightning
electric and magnetic field waveshapes and intensities at various distances, and to the
observed return-stroke speeds. Functions implied in these models are based either on the
double-exponential (DEXP) function (Bewley, 1929; Bruce & Golde, 1941) or Heidlers
function (Heidler, 1985). DEXP first order derivative at t=0
+
is too large, thus causing
numerical problems in LEMF calculations. It also has physically non-realistic convex
waveshape in the rising part. Heidlers function reproduces concave rising part and its
derivative is equal to zero at t=0+. It is also used for representing lightning currents in the
International standard IEC 62305 (IEC/TC 81, 2006). In order to obtain better agreement
with experimental results (Berger et al., 1975) one function was proposed (Nucci et al., 1990)
as a linear combination of the DEXP and Heidlers function. Another pulse function was
also proposed in (Feizhou & Shange, 2004). All these functions need peak correction factors,
and their parameters cannot be easily chosen according to the desired waveshape. However,
NCBC function parameters in the rising part can be chosen independently from parameters
in the decaying part. The exact rising time to the peak value can be chosen in advance, and
the decaying time to half of the peak value can be selected in the approximation procedure.
The maximum value of the function can be chosen without the peak correction factor.
Maximum current steepness can be adjusted (Javor & Rancic, 2011) using analytical
expression for the first derivative. Experimentally measured impulse charge and specific
energy values (Berger et al., 1975; Anderson & Eriksson, 1980) can be achieved using
analytically obtained integral of NCBC function and integral of the square of the function
(Javor, 2011a).
NCBC function (Fig. 1) is given with the following expression:

Fourier Transform Application in the Computation of Lightning Electromagnetic Field

59

( ) [ ]
( ) [ ]
1
/ exp (1 / ) , 0 ,
( )
/ exp (1 / ) , ,
i
a
m m m m
n
b
m i m i m m
i
I t t a t t t t
i t
I c t t b t t t t
=

(1)
where a and b
i
are parameters, c
i
coefficients, n the chosen number of expressions in the
decaying part, so that the total sum of n weighting coefficients c
i
is equal to unit, and t
m
is
the rise-time to the maximum current value I
m
. For n=1, c
1
=1 and b
1
=b, NCBC function
reduces to CBC function (Javor & Rancic, 2006) with four parameters (I
m
, t
m
, a and b). In the
special case, for n=1, a=4 and b=0.0312596735, CBC function reduces to High-Voltage Pulse
(HVP) function 1.2/50s (Velickovic & Aleksic, 1986). Impulse duration time is defined as
t
i
=t
k
-t
a
, for t
k
the time in which the current decreased to half of its peak value (Fig. 1). The
rising part of the function and its front rise-time are given in Fig. 2. The front rise-time is
defined as t
c
=t
b
-t
c
, for t
b
and t
c
determined as the time values corresponding to the points B
and A, obtained from intersecting horizontal lines for the maximum I
m
and the zero function
value (time axis) with the line drawn through the points A, for i(t)=0.3I
m
(Fig. 2) or for
i(t)=0.1I
m
in some other definitions, and B, for i(t)=0.9I
m
.

Fig. 1. Normalized NCBC function
NCBC function is an analytically prolonged mathematical function (but still continuous, so
as its first derivative, whereas higher order derivatives are not continuous at the point of
function maximum I
m
), the parameter a in the rising part can be chosen to approximate the
front of the waveshape independently from parameters b
i
and weighting coefficients c
i
in the
decaying part, which facilitates the approximation procedure. NCBC function belongs to C
1

differentiability class. Parameters of NCBC function can be chosen so that it represents
waveshape of the often used DEXP function with parameters given in (Bruce & Golde,
1941). DEXP function i(t)=I
m
[exp(-t)-exp(-t)] for I
m
=11kA, t
m
=0.5826s, =3
.
10
4
s
-1
, and
=10
7
s
-1
, has the decreasing time to half of the peak value of approximately 23s. It can be


60
approximated with NCBC function for n=1 and I
m
=11kA, t
m
=0.5826s, a=0.5 and b=0.02,
having also the decreasing time to half of the peak value of about 23s. If using lightning
stroke models and electromagnetic theory relations, lightning electric and magnetic field
components above perfectly conducting ground in general have three terms, depending on
the integral of the channel-base current function, on the function itself, and on the function
derivative. Consequently, the DEXP function having just a convex rising part and great
values of the first derivative at t=0
+
makes numerical problems in LEMF calculations.

Fig. 2. The rising part of the normalized CBC function representing current t
c
/t
i
=1.2/50s
NCBC function, for a > 1, satisfies the demand of having the first derivative equal to zero at
t=0
+
and the concave to convex rising part. The value of parameter a has to be chosen greater
than 1 in order to obtain one saddle point in the rising part (Javor & Rancic, 2006). Heidlers
function also has the first derivative equal to zero at t=0
+
, but it needs peak correction factor,
and its rise-time to the maximum current value cannot be chosen in advance. Heidler's
function doesn't have analytical integral and its Fourier transform cannot be obtained
analytically, but just numerically (Heidler & Cvetic, 2002).
NCBC function can approximate also other channel-base currents used in lightning return-
stroke modeling, as the one proposed in (Nucci et al., 1990). This function is used in many
papers and approximates well experimental results for subsequent negative strokes. In
order to represent the same current two terms are needed in the decaying part of NCBC
function (n=2) and its parameters are calculated as I
m
=11kA, t
m
=0.472s, a=1.1, b
1
=0.16,
c
1
=0.34, b
2
=0.0047, and c
2
=0.66.
2.1 Parameters of NCBC function
The normalized NCBC function for different values of parameter a, in the first 1s is
presented in Fig. 3, for t
m
=0.5826s. The parameter a determines the rising part of the


61
function. NCBC function is presented in Fig. 4 for different values of parameter b, for
t
m
=1.9s, but in a longer time period (300s), as parameter b determines the decaying part of
the function. If the rising time to the maximum value t
m
has values e.g. 0.5, 1, 2, 5, or 10s,
and other parameters are a=1.5 and b=0.01, the changes in the function waveshape are
presented in Fig. 5.

Fig. 3. Normalized NCBC function in the first 1s, for b = 0.03, t
m
=0.5826s, and different
values of a as parameter

Fig. 4. Normalized CBC function for a = 1.5, t
m
=1.9s, and different values of b as parameter


62

Fig. 5. Normalized CBC function for a = 1.5, b = 0.01, and different values of t
m
as parameter
2.2 Derivative of NCBC function
NCBC function first order derivative is:

( ) ( ) [ ]
( ) ( ) [ ]
1
1
1
/ 1 / exp (1 / ) , 0 ,
d ( )
d
/ 1 / exp (1 / ) , ,
i
a
m m m m m
n
b
m i i m m i m m
i
aI t t t t a t t t t
i t
t
I c b t t t t b t t t t

(2)
The first derivative of CBC function (NCBC for n=1) is presented in Fig. 6 up to 1s, for
b = 0.03 (but note that parameter b is irrelevant for the rising part), and different values of
a, t
m
and I
m
.
2.3 Integral of NCBC function
Integral of NCBC function is calculated as:

[ ]
( 1)
( 1)
0
( 1)
1
exp( ) ( 1, / ), 0
exp( ) ( 1, ) ( )d
exp( ) ( 1, / ) ( 1, ) ,
i
a
m m m m
t
a
m m
n
b
i i i i i m i i m
i
I t a a a at t t t
I t a a a a i t t
c b b b b t t b b t t
+
+
+
=
+ + =

+ +
(3)


63
Impulse charge is defined in IEC 62305 standard (IEC/TC 81, 2006) as the integral of the
channel-base current function. For standard lightning currents impulse charge is calculated
in (Javor, 2011a), so as specific energy as the integral of the square of NCBC function.

Fig. 6. The first derivative of CBC function in the first 1s
3. Fourier transform of NCBC function
Whether a function is given analytically, graphically or numerically, its Fourier transform
can be obtained analytically, numerically or using some of the commercial programs. Faster
or slower rising/decaying of the function and its waveshape at the end of the covered impulse
duration time determine the number N of needed points for Fast Fourier Transform.
For NCBC function Fourier and Laplace transforms are calculated analytically. It should be
noted that we obtain Fourier transform for any Fourier-transformable function which is zero
for t<0 if we substitute variable s with j2f in its Laplace transform.
The analytical expression for the unilateral Laplace transform of NCBC function is:

( ) ( )
1 1
1
exp( ) exp( )
( ) ( 1, ) ( 1, )
i
n
i i
m m m i i m
a b
i
m i m
c b a
I s I t a a st b b st
a st b st
+ +
=

= + + + + +

+ +

(4)
for
0
( 1, ) exp( )d
x
a
a x t t t + =
and ( 1, ) exp( )d
a
x
a x t t t

+ =
the incomplete Gamma functions,

as defined in (Abramowitz, Stegun, 1970).
Fourier transform of NCBC function is obtained from (4) for s= j2f, so:


64
( ) ( )
1 1
1
exp( ) exp( )
( ) ( 1, j 2 ) ( 1, j 2 )
j 2 j 2
i
n
i i
m m m i i m
a b
i
m i m
c b a
I f I t a a f t b b f t
a f t b f t

+ +
=

= + + + + +

+ +

. (5)
It can be obtained also numerically, or by using some computer program such as FAS
(Walker, 1996) with the application of FFT (Fast Fourier Transform). Any frequency domain
calculations can be done just for the positive frequencies, as for FFT of real functions the
following relations are valid:
{ } { } Re ( ) Re ( )
k N k
I f I f
= (6)
and
{ } { } Im ( ) Im ( )
k N k
I f I f
= . (7)
This feature makes the time for computing FFT twice shorter for real functions, which is
very useful if the number of FFT points is great so that calculations are time-consuming. For
smaller number of FFT points the computing time is shortened. FFT results given for NCBC
function representing lightning current indicate that the major part of its power is in the
lower frequency range. The modulus of FFT decays fast as the function of frequency, so it is
better to use some other than linear scaling of frequencies in order to cover the lower part of
the frequency range. For very high frequencies, the values of real and imaginary parts of
FFT of the pulse functions near the end of the frequency interval can be taken as zeros, as
they are relatively very small comparing to the values at lower frequencies. That doesn't
make much influence on computation results if IFFT (Inverse Fast Fourier Transform) is
done for obtaining time domain results based on frequency domain calculations. The
sufficient number of FFT points for lightning research studies should be 1024, but better
results are obtained with 2048 or 4096 points. This depends on the time interval to cover and
the current waveshape.
Fourier transform of Heidlers function is calculated approximately (Andreotti et al., 2005).
Some FFT results for Heidlers function are also given in (Vujevic & Lovric, 2010).
NCBC/CBC have analytical solutions for Fourier transform which enables obtaining
analytical solution for LEMF results in the case of perfectly conducting ground.
The modulus of Fourier transform of HVP function 1.2/50s is presented in Fig. 7, for I(f)
normalized to I(0), as the function of log
10
f. For f
2
6.4kHz10
3.8
Hz the normalized modulus
of FFT is |I(f
2
)/I(0)|0.3657. The results are obtained for FFT calculated in N=8192 points,
for the time step T19.064ns, frequency step f6.4kHz, and limit frequency f
g
52.455
MHz. From Fig. 7 is obvious that using frequencies higher than a few MHz is not necessary.
For some frequencies up to the chosen limit frequency the results are presented in Table 1
for FFT of HVP function 1.2/50s calculated in N=8192 points, for T19.064ns, f6.4kHz,
and f
g
52.455MHz. For the chosen number of points for FFT, the corresponding time
interval is T=NT156.17s and the frequency step is f=(NT)
-1
= T
-1
. For such values there
are 8192 frequencies in the chosen frequency interval [-0.5f
g
,+0.5 f
g
], and the highest positive
frequency is f
4097
26.2275MHz. FFT results for |I(f)|10
-n
|I(0)| for n=1, 2, 3, , 8, are also
given in Table 1, corresponding approximately to frequencies f
5
, f
35
, f
108
, f
352
, f
2478
, f
3912
, f
4078
,
f
4095
, and they also point out to the fast decaying of FFT modulus. As (6) and (7) are valid for


65
the real functions, this makes the time for computing twice shorter. This is very useful if the
number of FFT points is great and relevant calculations in frequency domain are time-
consuming. For smaller number of FFT points the total computing time is shortened, but the
sampling interval in time domain is critical in that case.
f
k
f(Hz) Re{I(f)} Im{I(f)} |I(f)/I(0)|
f
1
0.0000E+00 0.6232E-04 0.0000E+00 0.1000E+01
f
2
0.6403E+04 0.6892E-05 -0.2172E-04 0.3657E+00
f
3
0.1281E+05 0.8713E-06 -0.1179E-04 0.1896E+00
f
4
0.1921E+05 -0.2839E-06 -0.7867E-05 0.1263E+00
f
5
0.2561E+05 -0.6610E-06 -0.5842E-05 0.9433E-01
f
35
0.2177E+06 -0.6334E-06 -0.6164E-07 0.1021E-01
f
108
0.6851E+06 0.4510E-07 0.4597E-07 0.1033E-02
f
352
0.2248E+07 0.4931E-09 0.6259E-08 0.1007E-03
f
2478
0.1586E+08 0.8158E-12 0.6237E-09 0.1001E-04
f
3912
0.2503E+08 -0.2014E-12 0.6261E-10 0.1005E-05
f
4078
0.2611E+08 -0.1417E-12 0.6350E-11 0.1019E-06
f
4095
0.2621E+08 -0.1377E-12 0.6689E-12 0.1096E-07
f
4097
0.2623E+08 -1.3765E-13 0.0000E+00 2.2088E-09
Table 1. FFT of the HVP function in 8196 points

Fig. 7. Normalized FFT modulus of HVP as the function of frequency
For example, for N=2048 and the same time interval T=NT156.172s the sampling
frequency is f
g
=1/T13.114MHz, which gives the same frequency step f6.4kHz, but four
times greater step in time domain T76.256ns than if N=8092.


66
Fig. 8 shows real part of FFT for HVP function in the considered frequency range
[-4096f, +4096f], and enlarged rectangle area for the range of frequencies around zero
f[-200f, +200f]=[-1280.638kHz,+1280.638kHz]. It can be seen that Eq. (6) is valid.

Fig. 8. Real part of FFT for HVP in the frequency range [-0.5f
g
,+0.5f
g
]=[-4096f,+4096f] and
enlarged rectangle area for frequencies f[-200f ,+200f]
Fig. 9 shows imaginary part of FFT for HVP function in the same considered frequency
range [-0.5f
g
,+0.5f
g
], and enlarged rectangle area for f[-200f, +200f]. It can be seen that Eq.
(7) is valid.
FFT results given in Table 1 for this pulse function indicate that the major part of its power
is in the lower frequency range. The modulus of FFT decays rapid as the function of
frequency which can be noticed both in Table 1 and Fig. 7.

Fig. 9. Imaginary part of FFT for HVP in the range [-0.5f
g
,+0.5f
g
]=[-4096f,+4096f] and
enlarged rectangle area for frequencies f[-200f ,+200f]


67
As the value |I(f
2
)/I(0)|0.3657 is obtained already for the frequency f
2
6.4kHz (Fig. 7), it is
better to use some other then linear scaling of frequencies in order to cover better the lower
part of the frequency range. For very high frequencies, the values of real and imaginary part
of FFT near the end of the frequency interval can be taken as zeros, as they are relatively
very small comparing to their values at low frequencies. That doesnt make much influence
on the computation results if IFFT (Inverse Fast Fourier Transform) is done for obtaining
time domain results based on frequency domain calculations. Both FFT and IFFT can be
done using program FAS (Walker, 1996) in 128, 256, 512, 1024, 2048, 4096 or 8192 points. The
input data can be given as analytical functions (as in the case of NCBC or other channel-base
current functions) or discrete values in the corresponding set of values for that quantity (as
in the case of experimentally measured data).
In order to analyze lightning electromagnetic field in a far zone, as an external excitation of
some receiving antenna structure, vertical electric field and azimuthal magnetic field can be
represented with the waveshape of NCBC function according to the measured data and its
Fourier transform should be known. Fourier transform of the channel-base current is also
necessary if calculating equivalent voltage source as a product of the input impedance and
the channel-base current in frequency domain (Moini et al. 2000, Shoory et al. 2005), instead
of the current source itself at the channel-base. A few computer programs for antenna
analysis are used for calculations (Richmond, 1974, 1992; Bewensee, 1978; Burke & Poggio,
1980, 1981; Djordjevic et al., 2002), and most of these use the voltage generator as excitation
(Grcev et al., 2003).

Fig. 10. Normalized FFT modulus for NCBC, CBC and HVP as a function of frequency
For CBC function with parameters I
m
=11kA, t
m
=0.5826s, a=1.5 and b=0.02 the results of FFT
are presented in Figs. 10-12. This function is chosen adequate to the DEXP, except for the
parameter a = 1.5 instead of a = 0.5 in the rising part, in order to satisfy the mentioned


68
condition for lightning discharge channel-base current functions (having the first derivative
equal to zero at t=0
+
and the concave to convex shape in the rising part). In the same figures
there are the results for the new channel base current function (NCBC) with parameters
I
m
=11kA, t
m
=0.472s, a=1.1, b
1
=0.16, c
1
=0.34, b
2
=0.0047, and c
2
=0.66. Just for the comparison,

Fig. 11. Real part of the normalized FFT for NCBC, CBC and HVP functions in the frequency
range [-200f ,+200f]

Fig. 12. Imaginary part of the normalized FFT for NCBC, CBC and HVP functions in the
frequency range [-200f ,+200f]


69
the results are given in these figures also for the high voltage pulse (HVP) function 1.2/50s,
although the function with those parameters is not adequate for lightning channel-base
currents modeling. All the results are presented for the functions normalized to the
maximum values. The results for moduli of FFT for I(f) normalized to I(0), as the function of
frequency, are presented in Fig. 10. For f
2
6.4kHz the normalized modulus of FFT for NCBC
function is |I(f
2
)/I(0)|0.2426, for CBC function |I(f
2
)/I(0)|0.6242, and for HVP function
|I(f
2
)/I(0)|0.3657.
Real and imaginary parts of FFT in the frequency range f[-200f ,+200f] are presented in
Figs. 11 and 12.
Log-log dependence of FFT modulus on frequency is presented in Fig. 13 for CBC function
with parameters I
m
=11kA, t
m
=0.5826s, a=1.5 and b=0.02 and for NCBC function with
parameters I
m
=11kA, t
m
=0.472s, a=1.1, b
1
=0.16, c
1
=0.34, b
2
=0.0047, and c
2
=0.66.
Normalized FFT modulus as the function of frequency for NCBC with parameters
I
m
=25.176kA, t
m
=0.65s, a=20, b=0.00467, and for Heidlers function with parameters I
0
=25kA,
=0.993,
1
=0.454s,
2
=143s, both representing IEC 62305 standard negative first stroke
lightning current 0.25/100s for lightning protection level (LPL) III-IV, is given in Fig. 14.

Fig. 13. Normalized FFT moduli for NCBC and CBC functions as the function of frequency
The results show that for frequencies higher than 10MHz FFT modulus of these channel-
base current functions is less than 1
o
/
oo
of its value at 10kHz, so for frequencies higher than
10MHz calculations of LEMF are not needed in frequency domain.
FFT results for NCBC function given in Fig. 14 are also in good agreement with the results
given in (Vujevic & Lovric, 2010) for the corresponding Heidlers function with parameters
I
0
=25kA, =0.993,
1
=0.454s,
2
=143s, and for LPL III-IV.


70

Fig. 14. Normalized FFT moduli for NCBC and Heidlers functions representing standard
0.25/100s as the function of frequency
4. Electromagnetic field calculation using antenna model of the lightning
channel at a lossy ground
Based on experimentally measured characteristics of natural lightning (Berger et al., 1975;
Lin et al. 1979; Anderson & Eriksson 1980) and also artificially initiated lightning discharges,
LEMF can be estimated using some of the models from literature (Rakov & Uman, 1998,
2006), and lightning currents are assumed to propagate with attenuation and distortion
while distributing charge along the channel. In the case of perfectly conducting ground
calculations are simple in time domain, but full-wave approach in the case of a lossy ground
is complex to implement, so Fourier transform is applied. After calculations are done in
frequency domain, the conversion to time domain is performed by Inverse Fast Fourier
transform (IFFT). A solution of the Sommerfelds problem is required for the lossy ground.
A few alternatives are proposed in literature.
4.1 Alternatives to full-wave time domain computation
The problem of LEMF calculation can be easily solved directly in time domain if the ground
is treated as perfectly conductive. In such a case there exist vertical electric and azimuthal
magnetic field components in the observed point at the ground surface. Horizontal
component of electric field is zero at the perfectly conducting ground surface, but non-zero
above the ground surface. However, it exists above, under, and at the surface of the lossy
ground. Vertical component of the lightning electric and azimuthal component of the
magnetic field can be easily (but in that case approximately) determined at the distances
greater than a kilometer under the assumption of perfectly conducting ground. For smaller


71
distances, propagation above ground of finite conductivity results in the noticeable
attenuation of the high-frequency components of electric and magnetic field, and thus in
appearance of the horizontal electric field at the surface. Finite conductivity has greater
impact on horizontal than on vertical electric field, so calculation of horizontal component
requires rigorous computation or, at least, acceptable approximations.
Approximate formulas in frequency domain are often used for calculation of the horizontal
electric field in air, up to heights of tens of meters above the ground surface. These formulas
can be integrated in the calculation of LEMF in time domain, but the obtained expressions
are much more complex. There are simple approximations: the assumption of perfectly
conducting ground, "wavetilt" formula, Coorays approach and Rubinstein's approach.
Cooray, 1992, proposed the calculation of horizontal electric field at the surface of finitely
conductive ground using azimuthal magnetic induction and the expression for ground
surface impedance. He showed that this simple formula provides very accurate results at
the distances of about 200m. Rubinstein, 1996, proposed expression for the horizontal
electric field with two terms: 1) horizontal electric field calculated under the assumption of
perfectly conducting ground, and 2) the correction factor, given as a product of the magnetic
field calculated under the same assumption and the function similar as in wavetilt
formula which represents the effect of finite conductivity. The basic assumption of
Rubinstein's approximation is
1
>>
0
r1
, and that finite ground conductivity does not affect
the horizontal magnetic field at the surface. If this is not the case, then more general formula
can be written, known in literature as the Cooray-Rubinstein's formula (Cooray, 2002). Wait
gave generalization of Cooray-Rubinstein's formula and the exact evaluation of horizontal
electric field, showing under which circumstances this general expression reduces to
Cooray-Rubinstein's formula (Wait, 1997). Cooray and Lindquist, 1983, and Cooray, 1987,
using the attenuation function in time domain proposed by Wait, 1956, included effects of
the finite conductivity, and obtained results for the electric field that are in better agreement
with experiments. Terms for approximate formulas in time domain are complex, so the
approach in frequency domain is preferred.
4.2 Method of images
Method of images gives an approximate solution to the Sommerfelds integral. Complex
image technique often uses one or more terms of exponential series to approximate plane
wave reflection coefficient. Thus, multiple discrete and/or continuous image sources are
introduced, and this technique also proved not to be limited to a quasi-static range alone.
Using this technique, different authors (Shubair & Chow, 1993; Yang & Zhou, 2004; Popovic
& Petrovic, 1993; and Petrovic, 2005) obtained results for Sommerfelds integral kernel for
vertical dipoles above a lossy half-space which were used for the comparison with the new
Two-image approximation (TIA). Approximate formulas are often valid for a limited range
of ground electrical parameters, field point distances, or heights of dipoles above ground,
but TIA approximation of Sommerfelds integral kernel proposed in (Rancic & Javor, 2006 &
2007) has the advantage of being valid in a wide range of lossy ground electrical parameters,
for various heights of vertical electric dipoles above the ground, and for possibility to
calculate electromagnetic field in both near and far zone.
LEMF can be calculated using thin wire antenna modeling of a lightning channel assumed
as vertical, without branches and reflections from the end. If using electromagnetic model,


72
boundary conditions are fulfilled at the wire antenna surface, and a voltage or current
source is assumed at the channel-base. Unknown current distribution along the antenna is
determined by solving some system of integro-differential equations of EFIE or MFIE type
satisfying boundary condition for electric or magnetic field components, respectively. One
of those is e.g. System of integral equations of two potentials (SIE-TP), (Rancic, 1995), which
is of EFIE type. Current distribution along the antenna can be approximated e.g. by
polynomials (Popovic, 1970) with unknown complex coefficients. These can be determined
by using Method of Moments (MoM) (Harrington, 1968) i.e. by matching in a sufficient
number of points along the antenna. Based on that current distribution LEMF is calculated
using electromagnetic theory relations. Thus, the problem is solved in full (without
mentioned approximations of some LEMF components using others) on the basis of
approximate solution of Sommerfeld's problem of a dipole radiating at the arbitrary height
above the lossy half-space, which is a classical problem in electromagnetics.
The problem of a dipole radiation in the presence of a lossy medium was noticed yet in 1909,
by Sommerfeld, who determined the solution in the form of superposition of inhomogeneous
plane waves. His work had such a great influence on later theoretical research in this area
that the solution of more complex problems which shows the influence of a lossy medium
on the properties of linear antennas is marked as Sommerfeld's solution. Sommerfeld was
the first who treated all four cases of elementary Hertz's dipoles in the air applying his
formulation through the cylindrical waves. Van der Pol, 1931, proposed approximate
method for solving Sommerfeld's integrals. Class of integrals obtained by Sommerfeld's
formulation which strictly defines boundary conditions on the surface of discontinuity is
known in literature as integrals of Sommerfelds type. These integrals have limits of
integration 0 and infinity, and the integrand is a product of Bessel's function (J
0
or J
1
),
exponential, and one more function, so it is of very complex shape. Depending on the
complexity of the model, this function has different forms so that different problems appear
in its numerical integration. Since these integrals are highly oscillatory and slowly decreasing
functions they dont have solutions in the closed form. In the analysis of antennas above
lossy media, there is an inevitable and major problem to determine Sommerfeld's integral as
accurate as possible in a wide range of values of electrical parameters of the medium, for
different positions of current elements, different frequencies and positions of the observed
point in the field. There are many different ways to approximate Sommerfeld's integral
which can be found in literature. The model that includes complex mathematical functions
in the case of lossy ground gives result equivalent to the field of infinite number of current
images. Simpler approximations have limitations as for the values of electrical parameters of
the medium, the position of the point in the field (near or far zone) and positions of
matching points near or at the interface of two half-spaces. In order to include the influence
of electrical parameters of a real ground on antenna characteristics, direct Sommerfeld's
approach to the problem can be also used, but it is very complex for solving from both the
analytical and numerical point of view. In addition to the direct approach to this problem, it
is possible to modify the path of integration of Sommerfelds integral, or to use interpolation
and speed up the calculation, or to use tables of Sommerfeld's integrals, but there were also
some attempts to solve the problem by obtaining approximations in closed form for some
special cases of Sommerfeld's integrals. A suitable solution would be the one to determine
efficiently Sommerfeld's integrals for arbitrary positions and orientations of dipoles,
arbitrary positions of the points in LEMF and a wide range of frequencies.


73
One approximate solution is obtained by use of approximation of the spectral reflection
coefficient of a plane wave with exponential series, having one or more terms which is
named in literature as the method of images. For a horizontal electric dipole one complex
image is the simplest approximation, which was first introduced by Wait, 1955, and later
justified in (Wait & Spies, 1969). An important contribution was given by Bannister,
(Bannister, 1966), who also showed that techniques of the theory of images in the case of
finitely conductive ground are not restricted just to a quasi-static extent of the problem
(Bannister, 1978). He later published a review paper "Summary of image theory expressions
for the quasi-static fields of antennas at or above the earth's surface" (Bannister, 1979).
Bannister extended the validity of the approximation to high-frequency problems by
introducing exponential function exp(-u
0
d) where u
0
=(
2
+
0
2
)
1/2
and
0
is the complex
propagation constant in the air. Thus the effects of propagation are included, the same
solution obtained, and for both approximations there is the assumption that the refraction
coefficient of ground-to-air is much greater than 1 (n
10
=
1
/
0
>>1). Mahmoud & Metwally,
1981, used discrete, so as the combination of discrete and continuous images. Mahmoud,
1984, extended the theory to multiple images and to a layered ground. Distributed images
were also used and named the exact theory of continuous images in (Lindell & Alanen,
1984). By using Pronys method and nonlinear optimization technique Chow et al., 1991,
analytically expressed Sommerfeld's integrals in closed form of spatial complex images.
Shubair & Chow, 1993, discussed the impact of complex images on the problem of vertical
antennas in the presence of lossy media and took advantage of the spatial Green's function
in closed form in order to obtain a superposition of the sources influence, quasi-dynamic
image and three complex images. Arand et al., 2003, used the method of discrete complex
images and Generalized pencil of function technique to obtain the locations and impact of
current sources. A procedure to approximate Sommerfelds integrals was carried out by
Popovic & Petrovic, 1993, named "a simple near-exact solution" which uses a few images
determined so to approximately satisfy the boundary conditions in a limited area of the
interface and to yield accurate field values in the domain of the antenna. Electromagnetic
models including influence of the lossy ground on LEMF calculation often use
approximations of Sommerfelds integral kernel on the basis of theory of images in
frequency domain. Takashima et al., 1980, proposed a modified theory of images for smaller
distances from the point of observation to the vertical electric dipole. As in lightning
research is necessary to treat also larger distances from the point of observation to the
vertical electric dipole, this approximation can not satisfy the needs of computation in the
range of distances of interest. For larger distances from the point of observation to the
vertical electric dipole complete expressions, derived by King, 1990, were used for EM field
of a vertical electric dipole over the lossy medium, if the condition |
2
|
2
>>|
1
|
2
or
|
2
|>>3|
1
| is satisfied. King & Sandler, 1994, confirmed with their results the validity of
these expressions over certain types of ground.
The new approximation TIA can be classified into two-image approximations. The basic
idea for obtaining a new approximation of the spectral reflection coefficient is matching the
approximation of reflection coefficient as a two-terms function (with three unknown
constants) at two points, but also matching its first derivative at one point, which resulted in
a very efficient approximation of Sommerfeld's integral. TIA gives good results for
modeling in both near and far zone, as well as physically conceivable arrangement of
images and representation of the problem (as the distances of the images are real values and


74
not complex as in other approximations). TIA is similar to two-image approximations of
Sommerfeld's integral, and therefore is their abbreviation retained, but the way of deriving
the corresponding expressions is different. Sommerfeld's integral kernel results are
compared to the results from literature for different values of lossy ground electrical
parameters and various heights of vertical dipoles above the ground (Javor & Rancic, 2009).
For the spectral reflection coefficient the following approximation is used

0 0 0
( )
10 0
( ) e ,
c
u u d
z
R u B A

+
(8)
where A and B are unknown complex coefficients, d
0
is the distance from the source to the
second image, and u
0c
is the characteristic value chosen so that u
0c
=
0
. Unknown constants
are determined from equations obtained by matching the value of spectral reflection
coefficient in two points (u
0
and u
0
=
0
), and its first derivative at u
0
=
0
. This results in
B=R
, A=R
0
-R
and d
0
=
0
-1
(1+n
10
-2
), for R
and R
0
the quasi-stationary reflection
coefficients

2 2
10 10
( 1)/( 1) R n n
= + (9)
and

0 10 10
( 1) /( 1) R n n = + (10)
Instead of a complex distance of the second image the value is selected as d
im
=|
0
-1
(1+n
10
-2
)|,
based on numerical experiments.
4.3 Antenna model of a lightning discharge at a lossy ground
The simplest approximation of a LD channel is a vertical transmitting antenna, with an
excitation at its base, positioned at a lossy ground surface as in Fig. 15. For the application
of an electromagnetic model and equations of antenna theory, it is necessary to divide this
vertical antenna into segments of the length not greater than approximately half of the
wavelength corresponding to the frequency for which the analysis is done. Vertical rod
antenna with an excitation by an ideal Diracs voltage source in the channel-base, having
voltage u(t)=U(t) and frequency f, is presented in Fig. 15. The total height h of the antenna
models the height of a LD channel, which may be as high as several thousands of meters in
natural conditions. In such a model is assumed that the antenna is of a circular cross
section with the radius a, so that a<<
0
, where
0
is the wave-length in the air
corresponding to the frequency f. The corresponding angular frequency is =2f. The
antenna is divided into N segments, so that h=l
1
+ l
2
+...+l
N
, and the length of each segment
is l
k
>>a
k
, whereas radius a
k
<<
0
. It can be simply taken that a
k
=a for all segments, for k=1,
2,..., N. The division into segments is necessary because of the wide range of frequencies of
interest and for some of those the antenna should be divided into hundreds of segments.
The lossy ground is treated as homogeneous, linear and isotropic half-space of electrical
parameters: specific conductivity
1
, electric permittivity
1
=
0
r1
, and magnetic
permeability
1
=
0
. Other parameters of two half-spaces are defined as: the complex
conductivity
i
=
i
+j
i
, complex propagation constant
i
=(j
i
i
)
1/2
, for i=0 denoting air
and i=1 denoting ground of the relative complex permittivity
r1
=
r1
-j
i1
=
r1
-j60
1
0
, and
ground to the air refraction index n
10
=
1
/
0
=(
r1
)
1/2
.


75
Fig. 15. Lightning channel model at a lossy ground
Electric field is determined based on the current along the antenna from the expression

2
0

grad
z
E E E z
= = +
JG JJG
. (11)
Polynomial approximation used to represent the current distribution along k-th segment
with axis s
k
is given with:

, ,
0
( ) ( / ) ,
k
n
m
k k mk k k
m
I s B s l
=
=

(12)
where n
k
is the polynomial degree and B
mk
are unknown complex coefficients to be
determined from the system of equations SIE-TP. By satisfying boundary condition for the
tangential component of electric field, SIE-TP is obtained as:
[ ]
2
0 0 0 0
0
0 0
( )ch ( ) d ( )sh ( ) d
n n
n
s s
n s n
s s
s s s s s s s s
= =

+

[ ]
0 2 0
0
'( ) ( )sh ( ) d sh( ),
n
s
n n n n n
s
Z s I s s s s C s
=
+ =
(13)
[ ] [ ]
2
0 0 0 0 0 0
0 0
( ) ( )sh ( ) d ( )ch ( ) d
n n
n
s s
n n s n
s s
s s s s s s s s s
= =
+ +

[ ]
0 2 0
0
'( ) ( )ch ( ) d ch( )
n
s
n n n n n
s
Z s I s s s s C s
=
+ =
. (14)
Polynomial approximation is used to represent the current distribution along k-th segment
with axis s
k
, so that in (13) and (14), (s)=-div
JJG
0
=-/z
0
represents the scalar potential at


76
the antenna surface,
sn
(s)=
z0
(s) Hertz vector tangential component, whereas Z
n
(s) is the
impedance per unit length, s
n
the matching point, and s the local coordinate along the n-th
conductor, so that 0s l
n
.
z0
(s) and
0
(s) are potentials in the upper half-space to be
calculated from:

0 0 1 00 2
1 0
' 0
1
( ) ( )[ ( ) ( )]d
4
k
k
l
N
v
z k k k k k
k
s
s I s K r S r s
=
=
= +

, (15)

0 0 1 00 2
1 0
' 0
1
( ) ( ) [ ( ) ( )]d
4
k
k
l
N
v
k k k k k
k k
s
s I s K r S r s
s
=
=

, (16)
for K
0
(r
1k
)=exp(-
0
r
1k
)/r
1k
the standard potential kernel, and S
00
(r
2k
) the Sommerfelds
integral kernel defined as

00 2 10 0 2
0
( ) ( ) ( , )d .
v
k z k
S r R K r
=
=

(17)
10
( )
z
R
is the reflection coefficient in the spectral domain, for the variable 0<, so that

2
10 0 1
10 10 0
2
10 0 1
( ) ( ) ,
z z
n u u
R R u
n u u
= =
+

(18)
for
2 2 2 2 2
1 1 0 1 0
, u u = + = + and
2 2
0 0
, u = + whereas
0 2
( , )
k
K r
is the spectral
form of the standard potential kernel

0
( ')
0 2 0
0
e
( , ) ( )
k
u z s
k
K r J
u
+
=
(19)
and

0 2
0 2 0 2
2
0
( ) ( , )d ,
k
r
k k
k
e
K r K r
r

=
= =

(20)
for
2 2
2
( ')
k k
r z s = + + the distance from the field point M (, , z) to the first image of the
k-th antenna segment for
2 2
x y = + , and
0
( ) J the Bessel function of the first kind and
order zero.
Thus, for the Sommerfelds integral kernel (7) is obtained:

00 2 0 2 0 0 3
( ) ( ) exp( ) ( ),
v
k k im k
S r BK r A d K r + (21)
for
2 2
3
( ' )
k k im
r z s d = + + + the distance from the second image to the field point, and
0 3 0 3 3
( ) exp( ) /
k k k
K r r r = .
LEMF calculations were also done for the antenna models of cage structures (Javor, 2003)
and in the case of lightning protection rods at a lossy ground (Javor & Rancic, 2009).


77
4.4 Current distribution along the vertical antenna at the lossy ground
The results obtained with TIA approximation are shown in Fig. 16 for real and imaginary
part of the current along the vertical antenna having height h=300m, radius a=0.05m, and
resistivity per unit length R=0.1/m, at the lossy ground of relative dielectric constant
r1
=2 and
r1
=10, specific conductivity
1
=10
-1
S/m or 10
-5
S/m, for frequency f=3MHz, the
number of antenna segments N=30, and the degree of polynomial approximation n
k
=3, for
k=1,...,30.

Fig. 16. Real and imaginary part of the current along the antenna for h=300m and radius
a=0.05m, for different lossy ground parameters and f=3MHz
4.5 Input impedance of a vertical antenna modeling the lightning discharge channel
Input impedance, Z
ul
, is important for the antenna analysis in frequency domain and
presents integral characteristic of the antenna structure, which also allows checking the
accuracy of TIA for Sommerfeld's integral kernel. Satisfactory agreement for the polynomial
degree n>2 was confirmed. It is enough to choose a polynomial degree n=2 if the length of
the antenna is not greater than 0.6
0
, for N=1 segment of the antenna. The polynomial
degree should not take values n>8 as the polynomial approximation of the antenna current
is not appropriate for those. For
1
0
<10
-1
the obtained values for the input
impedance/admittance are not dependent on the normalized conductivity, but
approximately equal to the values of the input impedance/admittance in the case of a
perfect dielectric of relative dielectric constant
r1
.
For the antenna of length h=2600m the results for input impedance are obtained for the
frequency step f =6.425kHz and the selected maximum frequency f
max
/2=3.2896MHz for
FFT transforming time interval [0,t] into the frequency interval [-f
max
/2, f
max
/2]. For an
arbitrary overall height h the segmentation should be done depending on the frequency i.e.
wavelength into the segments of length l
k

0
/2 for a selected polynomial degree n
k
=3. For
frequencies f <500kHz the antenna can be treated as one segment, so that N=1 is enough for
calculations, whereas for higher frequencies is necessary to divide the antenna into
segments. E.g. about 20 segments are required for frequencies around 1MHz, and about 200
segments for frequencies around 10MHz, if the chosen polynomial degree of the current
approximation is n
k
=3 along each of the segments. Fig. 17 shows the results for input
resistance and input reactance of the antenna modeling lightning discharge channel, for


78
h=2600m, a=5cm, Z=0.1 /m,
r1
=10 and
1
=0.01S/m, N
FFT
=1024 points for FFT and
f=6.425kHz. Fig. 18 shows the results for input conductance and input susceptance. For
different frequencies, different number of segments along the VMA in the range 1N200
was chosen, depending on the observed frequency.

Fig. 17. Input resistance and reactance of the vertical antenna, for h=2600m, a=0.05m, and
Z=0.1/m, at the lossy ground of parameters
r1
=10 and
1
=0.01S/m, versus frequency

Fig. 18. Input conductance and susceptance of the vertical antenna, for h=2600m, a=0.05m,
and Z=0.1/m, at the lossy ground of parameters
r1
=10 and
1
=0.01S/m, versus frequency
4.6 Frequency domain results for lightning electric field
Results for vertical and radial electric field as functions of the normalized radial distance
0
for the normalized antenna height h/
0
=0.25, normalized radius of the antenna
a/
0
=0.0005 at the ground of specific conductivity
1
=0.01S/m and for different values of
electric permittivity
r1
as parameter, are presented in Fig. 19, and for
r1
=10 and for
different values of specific conductivity
1
as parameter in Fig. 20. Vertical and radial
component of the electric field at radial distances r from 50 to 250m (0.5
0
r2.5
0
) from


79

Fig. 19. Vertical and radial electric field of the quarter-wavelength monopole antenna at the
lossy ground, for
1
=0.01S/m and
r1
as parameter, as functions of normalized distance

Fig. 20. Vertical and radial electric field of the quarter-wavelength monopole antenna at the
lossy ground, for
r1
=10 and
1
as parameter, as the functions of normalized distance
the base of the antenna having height h=300m, circular cross section of radius a=5cm at
the ground surface of parameters
r1
=10 and
1
=0.01S/m, for the frequency f=3MHz, the
polynomial degree of the current distribution approximation along each of the segments
n
k
=3, and the number of segments 20, 30 and 50, is presented in Fig. 21. The results
obtained for the electric field in the points at a height z=1.5m above the ground surface,
for the radial distances 0.5
0
r2.5
0
, differ a little from results for the field at the ground
surface (z=0).
4.7 Time domain results for lightning electromagnetic field
Results for vertical electric and azimuthal magnetic field components are presented in Figs.
22-24 for different distances from the channel-base: r=500m, 5km and 100km. For CBC


80

Fig. 21. Vertical and radial electric field of the antenna h=300m at the lossy ground, for
r1
=10 and
1
=0.01S/m for f=3MHz, as the functions of distance

Fig. 22. Vertical electric and azimuthal magnetic field at the ground surface (z=0) for r=500m
function with parameters I
m
=11kA, t
m
=0.5826s, a=1.5 and b=0.02, and for NCBC function
with parameters I
m
=11kA, t
m
=0.472s, a=1.1, b
1
=0.16, c
1
=0.34, b
2
=0.0047, and c
2
=0.66, and
two different decaying constants =2000m and =4500m, the results are compared to the
results from (Nucci, 1990) calculated for perfectly conducting ground using the same
Modified Transmission Line Model with Exponential Decay (MTLE) with the decaying
constant =2000m (Nucci et al., 1990), the same return-stroke speed v=1.3
.
10
8
m/s and the
channel having height H=2600m and radius a=0.05m. For the same v, H and a, for the
distributed resistance R=0.1/m along the antenna, driven by a Dirac delta current source
connected across a 3.25m gap, the results are presented also for the perfectly conducting
ground (Shoory et al., 2005). Shoory et al., 2005, used the electromagnetic model and a
similar procedure to here presented: EFIE type equation, Method of Moments (MoM),
method of images for the approximate solution of Sommerfelds integrals (another


81
approximation), concept of current source, NEC2 computer program (Burke & Poggio,
1981) and FFT/IFFT in 8192 points, i.e. calculations for 4092 positive frequencies. It is
interesting that results very similar to (Shoory et al., 2005) would be obtained with NCBC
& MTLE, but for =6000m.

Fig. 23. Vertical electric and azimuthal magnetic field at the ground surface (z=0) for r=5km

Fig. 24. Vertical electric and azimuthal magnetic field at the ground surface (z=0) for
r=100km
5. Conclusion
Fourier transform proved to be very successful in lightning research. It enables calculations
in frequency domain which are more suitable for including lossy ground effects than in
time domain. It also provides information about frequency spectra of the quantities of
interest in lightning research. For antenna modeling of a lightning stroke in frequency
domain, Sommerfelds integral is calculated efficiently using Two-image approximation.


82
Results are in good agreement with the results from literature for various ground electrical
parameters, heights of vertical dipoles above ground, in the near and far field. Based on
these results, it can be concluded that the effects of lossy ground are greater on horizontal
than on vertical electric field, and that specific conductivity influences more than electrical
permittivity. Fourier transform application has to be further investigated in terms of
optimal choice of FFT parameters in order to reduce computing time which can be
important in antenna modeling of lightning discharge channels and in analysis of lightning
electromagnetic fields.
6. Acknowledgment
This work is supported by Project Grant III44004 (2011-2014) financed by Ministry of
Education and Science, Republic of Serbia.
7. References
Abramowitz, M. & Stegun, I. (1970). Handbook of Mathematical Functions, pp. 253-256, Dover
Publications, ISBN 486-61272-4, New York, USA
Anderson, R. B. & Eriksson, A. J. (1980). Lightning parameters for engineering application,
Electra, Vol.69, pp. 65-102
Andreotti, A.; Falco, S. & Verolino L. (2005). Some integrals involving Heidlers lightning
return stroke current expression, Electrical Engineering (Archiv fr Elektrotechnik),
No.87, pp. 121-128, ISSN 0948-7921
Arand, B. A.; Hakkak M.; Forooraghi K. & Mohassel, J. R. (2003). Analysis of vertical wire
antenna above lossy ground using discrete complex image method, Int. Journal of
Electronics and Communications (AE), Vol.57, No.5, 2003, pp. 333-337
Baba, Y. & Ishii, M. (2003). Characteristics of electromagnetic return-stroke models, IEEE
Trans. on Electromagn. Compat., Vol.45, No.1, (Feb. 2003), pp.129-135
Bannister, P. R. (1966). Quasi-static Fields of Dipole Antennas at the Earths Surface, Radio
Science, Vol.1, No.11, (November 1966), pp. 1321-1330
Bannister, P. R. (1978). Extension of quasi-static range finitely conducting earth-image
theory techniques to other ranges, IEEE Trans. on Antennas and Propagation, Vol.AP-
26, No.3, (May 1978), pp. 507-508
Berger, K.; Anderson, R. B. & Krninger, H. (1975). Parameters of lightning flashes, Electra,
Vol.41, pp. 22-37
Bewensee, R. M. (1978). WF-SYR/LLL1: A thin-wire computer code for antennas and
scatterers with pulse expansion functions for currents, Lawrence Livermore
Laboratory, CA, Rep. UCRL 52 028
Bewley, L. V. (1929) Traveling waves due to lightning, AIEEE Transactions, Vol. 48, pp. 1050-
1064.
Bruce, C. E. R. & Golde, R. H. (1941). The lightning discharge, J. Inst. Electr. Eng., Vol.88, pp.
487-520, ISSN 0360-6449
Burke, G. J. & Poggio, A. J. (1980, 1981). Numerical Electromagnetics Code (NEC) - Method
of Moments, Technical Document 116, Naval Ocean Systems Center, San Diego,
California


83
Chow, Y. L.; Yang, J. J.; Fang, D. G. & Howard G. E. (1991). A closed-form spatial Greens
function for the thick microstrip substrate, IEEE Trans. on Microwave theory and
techniques, Vol.39, No.3, (Mar. 1991), pp. 588-592
Cooray, V. (1987). Effects of propagation on the return stroke radiation fields, Radio Sci.,
Vol.22, pp. 757-768
Cooray, V. (1992). Horizontal fields generated by return strokes, Radio Sci., Vol.27, pp. 529-
537
Cooray, V. (2002). Some considerations on Cooray-Rubinstein formulation used in deriving
the horizontal electric field of lightning return strokes over finitely conducting
ground, IEEE Trans. on Electromagn. Compat., Vol.44, No.4, (Nov. 2002), pp. 560-566
Djordjevic, A. R.; Bazdar, M. B.; Petrovic, V. V.; Olcan, D. I.; Sarkar, T. K. & Harrington, R. F.
(2002). AWAS for Windows, Version 2.0, Analysis of wire antennas and scatterers -
Software and User's manual, Artech House
Feizhou, Z. & Shange L. (2002). A new function to represent the lightning return-stroke
currents, IEEE Trans. on Electromagn. Compat., Vol.44, No.4, (Nov. 2002), pp. 595-597
Golde, R. H. (1977). Lightning currents and related parameters, In : Lightning, Vol. 1, Physics
of Lightning, R. H. Golde, (Ed.), pp. 309-350, Academic Press, ISBN 0122878027,
London, UK
Grcev, L.; Rachidi, F. & Rakov, V. A. (2003). Comparison of electromagnetic models of
lightning return strokes using current and voltage sources, International Conference
on Atmospheric Electricity, ICAE'03, (June 2003), Versailles, France
Harrington, R. F. (1968). Field Computation by Moment Methods, New York: Macmillan, New
York, Sec. 6.2.
Heidler, F. & Cvetic, J. (2002). A class of analytical functions to study the lightning effects
associated with the current front, ETEP, Vol.12, No.2, (March/April 2002), pp. 141-
150, ISSN 1546-3109
Heidler, F. (1985). Travelling current source model for LEMP calculation, Proc. 6
th
Int. Zurich
Symp. EMC, pp. 157-162, Zurich, Switzerland, March 1985
IEC 62305-1 Ed. 1.0 (2006). Protection against lightning - Part 1: General principles,
International Standard, IEC TC/SC 81
Javor, V. & Rancic, P. D. (2006). Application of One Suitable Lightning Return Stroke
Current Model, Proceedings of the International Symposium on Electromagnetic
Compatibility EMC Europe 2006, pp. 941-946, Barcelona, Spain, September 4-8, 2006
Javor, V. & Rancic, P. D. (2007). Vertical Monopole Antenna Model of the Lightning
Discharge Current: Two-Image Approximation of Sommerfeld's Integral Kernel,
CD Proc. of the 16
th
Conf. COMPUMAG 2007, Aachen, June 2007
Javor, V. & Rancic, P. D. (2009). Electromagnetic field in the vicinity of lightning protection
rods at a lossy ground, IEEE Trans. on Electromagn. Compat., Vol.51, No.2, (May
2009), pp.320-330, ISSN 0018-9375
Javor, V. & Rancic, P. D. (2011). A Channel-Base Current function for lightning return-stroke
modeling, IEEE Transactions on Electromagn. Compat., Vol.53, No.1, (February 2011),
pp. 254-259, ISSN 0018-9375


84
Javor, V. (2003). Electromagnetic Field Distribution of a nearby Lightning Discharge inside
an Overground Conductive Wire Structure, Facta Universitatis, Series Electronics and
Energetics, Nis, Vol.16, (Apr. 2003), pp. 41-53, ISSN 0353-3670
Javor, V. (2008). Approximating Decaying Part of the Lightning Return Stroke Channel-Base
Current, Proceedings of the 3
rd
International Symposium on Lightning Physics and
Effects, pp. 26, Vienna, Austria, April 14-15, 2008
Javor, V. (2011a). Impulse charge and specific energy of lightning discharge currents, CD
Proceedings of the 10
th
International Conference on Applied Electromagnetics, PES 2011,
Paper O1_1, (September 2011), Nis, Serbia
Javor, V. (2011b). Approximation of a lightning channel-base current with a two-rise front,
The Int. Journal for computation and mathematics in electric and electronic engineering
COMPEL, ISSN 0332-1649, (accepted for publication), 2011
Javor, V. (2011c). New functions for representing IEC 62305 standard and other typical
lightning currents, Journal of lightning research, ISSN 1652-8034, Ref. BSP-JLR-2011-4,
(accepted for publication), 2011
Lin, Y. T.; Uman, M. A.; Tiller, J. A.; Brantley R. D.; Beasley, W. H.; Krieder, E. P. &
Weidman, C. D. (1979). Characterization of lightning return stroke electric and
magnetic fields from simultaneous two-station measurements, J. Geophys. Res.,
Vol.84, pp. 6307-6314, ISSN 0148-0227
Lindell, I. V. & Alanen, E. (1984). Exact Image Theory for the Sommerfeld Half-Space
Problem, Part III: General Formulation, IEEE Trans. on Ant. and Prop., Vol.32, No.10,
pp. 1027-1032
Mahmoud, S. F. & Metwally, A. D. (1981). New image representation for dipoles near a
dissipative earth: Part 1. Discrete images; Part 2. Discrete plus continuous images,
Radio Science, Vol.16, No. 6, (Nov-Dec. 1981), pp. 1271-1283
Mahmoud, S. F. (1984). Image theory for electric dipoles above a conducting anisotropic
earth, IEEE Trans. on Antennas and Propagation, Vol.AP-32, No.7, (July 1984), pp.
679-683
Moini, R.; Kordi, B.; Rafi, G. Z. & Rakov, V. A. (2000). A new lightning return stroke model
based on antenna theory, Journal of Geophys. Research, Vol.105, No.D24, pp. 29693-
29702, ISSN 0148-0227
Nucci, C. A.; Diendorfer, G.; Uman, M. A.; Rachidi, F.; Ianoz, M. & Mazzetti, C. (1990).
Lightning return stroke current models with specified channel-base current: a
review and comparison, Journal of Geophys. Research, Vol.95, No.D12, (November
1990), pp. 20395-20408, ISSN 0148-0227
Petrovic, V. V. (2005). Private communication, Department of Theoretical Electrical Eng.,
ETF-School of EE, University of Belgrade
Popovic, B. D. & Petrovic, V. V. (1993). Vertical wire antenna above ground: simple near-
exact image solution, IEE Proceedings-H, Vol.140, No.6, (Dec. 1993), pp. 501-507
Popovic, B. D. (1970). Polynomial approximation of current along thin symmetrical
cylindrical dipoles, Proceedings of IEEE, Vol.117, No.5, (May 1970), pp. 873-878
Rakov, V. A. & Uman, M. A. (1998). Review and evaluation of lightning return stroke
models including some aspects of their application, IEEE Trans. on Electromagn.
Compat., Vol.40, No.4, (November 1998), pp. 403-426, ISSN 0018-9375


85
Rakov, V. A. & Uman, M. A. (2006). Lightning Physics and Effects, pp. 394-431, Cambridge
University Press, ISBN 978-0-521-58327-5, Cambridge, UK
Rancic, M. P. & Rancic, P. D. (2006). Vertical dipole antenna above a lossy half space:
Efficient and accurate two-image approximation for the Sommerfelds integral, CD
Proc. of EuCAP 2006, paper No.121 (Ref. No. 362128), Nice, Nov. 2006
Rancic, P. D. & Javor, V. (2007). New Two-Image Approximation of Sommerfeld's Integral
Kernel in Calculation of Electromagnetic Field of Vertical Mast Antenna in
Frequency Domain, 2
nd
Int. Symp. on Lightning Physics and Effects, (April 2007),
European COST Action P-18, Vienna, Austria
Rancic, P. D. (1995). Contribution to linear antennas analysis by new forms of integral
equations of two potentials, Proc. of 10
th
Conf. COMPUMAG'95, Berlin, (July 1995),
pp. 328-329
Richmond, J. H. (1974). Computer program for thin-wire structures in a homogeneous
conducting medium, NASA Report CR-2399, National Technical Information
Service, Springfield, Virginia
Richmond, J. H. (1992). Radiation and scattering by thin-wire structures in the complex
frequency domain, In: Computational Electromagnetics, E. K. Miller (Ed.), IEEE Press,
New York, USA
Rubinstein, M. (1996). An approximate formula for the calculation of the horizontal electric
field from lightning at close, intermediate, and long range, IEEE Trans. on
Electromagn. Compat., Vol.38, No.3, (August 1996), pp. 531-535
Shoory, A.; Moini, R.; Sadeghi, H. & Rakov, V. A. (2005). Analysis of lightning-radiated
electromagnetic fields in the vicinity of lossy ground, IEEE Trans. on Electromagn.
Compat., Vol.47, No.1, (February 2005), pp. 131-145, ISSN 0018-9375
Shubair, R. M. & Chow, Y. L. (1993). A simple and accurate complex image interpretation of
vertical antennas present in contiguous dielectric half-spaces, IEEE Trans. on
Antennas and Propagation, Vol.41, No.6, (June 1993), pp. 806-812
Sommerfeld, A. N. (1909). ber die Ausbreitung der Wellen in der Drahtlosen Telegraphie,
Ann. der Physik, Vol.28, pp. 665-736
Takashima, T., Nakae, T. & Ishibashi R. (1980). Calculation of complex fields in conducting
media, IEEE Trans. on Electr. Insul., Vol.EI-15, pp. 1-7
Van der Pol, B. (1931). ber die Ausbreitung elektromagnetischer Wellen, Z.
Hochfrequenctechnik 87, pp. 152-157
Velickovic, D. M. & Aleksic, S. R. (1986). A new approximation of pulse phenomena Proc. II
Serbian Symp. Applied Electrostatics PES'86, Nis, Serbia, (Nov. 1986), pp. 6.1-6.9, (in
Serbian)
Vujevic, S. & Lovric, D. (2010). Exponential approximation of the Heidler function for the
reproduction of lightning current waveshapes, Electric Power Systems Research,
Vol.80, No.10, (October 2010), pp. 1293-1298, ISSN 0378-7796
Wait, J. R. & Spies, K. P. (1969). On the Image Theory Representation of the Quasi-static
Fields of a Line Source Above the Ground, Can. J. Phys., No.47, pp. 2731-2733
Wait, J. R. (1997). Concerning the horizontal electric field of lightning, IEEE Trans. on
Electromagn. Compat., Vol.39, No.2, (1997), pp. 186


86
Walker, J. S. (1996). Fast Fourier Transforms, pp. 149-169, Boca Raton: CRC Press, ISBN
0-8493-7163-5
Yang, C. & Zhou, B. (2004). Calculation methods of electromagnetic fields very close to
lightning, IEEE Trans. on Electromagn. Compat., Vol.46, No.1, (Feb. 2004), pp. 133-141
4
Robust Beamforming and DOA Estimation
Liu Congfeng
Electronic Countermeasure Research Institute, Xidian University, Xian Shaanxi,
China
1. Introduction
1.1 Robust beamforming overview
Beamforming is a ubiquitous task in array signal processing with applications, among others,
in radar, sonar, acoustics, astronomy, seismology, communications, and medical imaging.
Without loss of generality, we consider herein beamforming in array processing applications.
The introduction to beamforming can be found in [1]-[9] and the references herein.
The traditional approach to the design of adaptive beamformers assumes that the desired
signal components are not present in training data, and the robustness of beamformer is
known to depend essentially on the availability of signal-free training data. However, in
many important applications such as mobile communications, passive location, microphone
array speech processing, medical imaging, and radio astronomy, the signal-free training
data cells are unavailable. In such scenarios, the desired signal is always present in the
training snapshots, and the adaptive beamforming methods become very sensitive to any
violation of underlying assumptions on the environment, sources, or sensor array. In fact,
the performances of the existing adaptive array algorithms are known to degrade
substantially in the presence of even slight mismatches between the actual and presumed
array responses to the desired signal [10]-[12]. Similar types of degradation can take place
when the array response is known precisely but the training sample size is small, namely
the mismatch between the actual and the estimated covariance matrix [13]-[15]. Therefore,
robust approaches to adaptive beamforming appear to be of primary importance in these
cases [16][17].
Many approaches have been proposed to improve the robustness of the adaptive
beamformer during the past three decades. Indeed, the literatures on the robust adaptive
beamformer are quite extensive. We provide a brief review as fellows. For more recent
detailed critical reviews, see [18] and [19]-[25].
1.1.1 Robust approaches for signal direction mismatch
For the specific case of the signal direction mismatch, there are several efficient methods
have been developed. Representative examples of such techniques are the linearly
constrained minimum variance (LCMV) beamformer [26], which is also denoted as the
linearly constrained minimum power (LCMP) beamformer in other references [27] and this
chapter, signal blocking-based algorithms [10][28], and Bayesian beamformer [29]. Although


88
all these methods provide excellent robustness against the signal direction mismatch, they
are not robust against other types of mismatches caused by poor array calibration, unknown
sensor mutual coupling, near-far wavefront mismodeling, signal wavefront distortions,
source spreading, and coherent/incoherent local scattering, as well as other effects [17].
Chun-Yang Chen and P.P.Vaidyanathan consider a simplified uncertainty set which
contains only the steering vectors with a desired uncertainty range of direction of arrival
(DOA) [25], although the closed-form solution is given, and the diagonal loading level can
be computed by the iteration method systematically, but how to determine the DOA
uncertainty range is the critical problem.
1.1.2 Robust approaches for general mismatch
Several other approaches are known to provide the improved robustness against more
general types of mismatches, for example, the algorithms that use the diagonal loading of
the sample covariance matrix [14][16], the eigenspace-based beamformer [11][30][31], and
the covariance matrix taper (CMT) approach [32]-[34]. For the diagonal loading method, a
serious drawback is that there is no reliable way to choose the diagonal loading level,
F.Vincent and O.Besson propose the method to select the optimal loading level with a view
to maximizing the signal-to-noise ratio (SNR) in the presence of steering vector errors and it
is shown that the loading is negative, but they cant give the exact solution, instead of the
approximate solution, moreover, they cant give the expression of steering vector errors [35].
The eigenspace-based approach is essentially restricted in its performance at low SNR and
when the dimension of the signal-plus-interference subspace is high, and the dimension
must be known in the latter technique [31]. The CMT approach is known to provide an
excellent robustness in scenarios with nonstationary interferers, however, its robustness
against mismatches of the desired signal array response may be unsatisfactory, furthermore,
it can also be explained as the diagonal loading [33].
1.1.3 Uncertainty set constraint approaches for general mismatch
Very recently, many approaches have been proposed for improving the robustness of the
standard minimum variance distortionless response (MVDR) beamformer. Their main ideas
are based on the definition of the uncertainty set and the worst-case performance
optimization, but these algorithms are all classified to the diagonal loading technique.
Jian Li et al propose the robust Capon beamformer under the constraint of steering vector
uncertainty set [20], then the constraint of steering vector norm is imposed and the doubly
constraint robust Capon beamformer is proposed [22]. For the two beamformers, although
they give the exact weight vectors, and the methods of finding the optimal loading level, but
their performance improvements are not obvious. Actually, the constraint of uncertainty set
is the essence of the two robust beamformers, and the two beamfomer have the same
robustness characteristic. F.Vincent and O.Besson also analyze the performance of the
beamformer under the uncertainty set constraint approximatively, but they cant give the
exact loading level [36].
S. A. Vorobyov et al propose a robust beamformer in the presence of an arbitrary unknown
signal steering vector mismatch [19], although they prove the proposed approach equivalent
to the loading sample matrix inversion (LSMI) algorithm, but they cant give the direct


89
method to compute the optimal weight vector, and the second-order cone (SOC)
programming-based approach is used to solve the original problem. Ayman Elnashar et al
make use of the diagonal loading technique to implement the robust beamformer [24], but
the optimal value of diagonal loading level is not solved exactly, alternatively, the diagonal
loading technique is integrated into the adaptive update schemes by means of optimum
variable loading technique. R. G. Lorenz and S. P. Boyd also solve the similar beamformer
by the Lagrange multiplier techniques [23], but they express the weight vector and the array
manifold as the direct sum of the corresponding real and imaginary components. Almir
Mutapcic et al show that worst-case robust beamforming with multiplicative uncertainty in
the weights can be cast as a tractable convex optimization problem [37], but they cant give
the solving method, In fact, the proposed robust beamformer with uncertain weights can be
converted to that in [19] equivalently.
S. Shahbazpanahi et al consider the general-rank signal model, and the robust beamformer
is proposed for the distributed sources [21], therein, an elegant closed-form solution is
given, but its performance improvement depends on the constraint parameter severely, and
is not up to optimal.
1.1.4 Weight norm constraint approaches for general mismatch
Jian Li et al propose a Capon beamforming approach with the norm inequality constraint
(NIC) to improve the robustness against array steering vector errors and noise [22], although
the exact solution is given, and optimal loading level can be computed via the proposed
method, but by analysis and simulation, its efficiency is not as good as expectation. Since the
constraint parameter determines its robustness, but how to select the constraint parameter is
not discussed.
Quadratic inequality constraints (QIC) on the weight vector of LCMP beamformer can
improve robustness to pointing errors and random perturbations in sensor parameter [27].
The weights that minimize the output power subject to linear constraints and an inequality
constraint on the norm of the weight vector have the same form as that of the optimum
LCMP beamformer with diagonal loading of the data covariance matrix. But the optimal
loading level cannot be directly expressed as a function of the constraint in a closed form,
and cannot be solved exactly. Hence, its application is restricted by the optimal weight
vector finding. So that some numerically algorithms are proposed to implement the
QICLCMP, such as Least Mean Squares (LMS) or Recursive Least Squares (RLS) [27], but the
application effect isnt good as the expectation.
This chapter is organized as follows [38]. First, the norm inequality constraint Capon
beamformer (NICCB) is introduced and analyzed particularly. Second, the choice of the
norm constraint parameter and the selecting bound is discussed. Third, the norm equality
constraint Capon beamformer (NECCB) is proposed and is solved effectively. Finally, the
simulation analyses and the conclusion are given.
1.2 Capon beamformer under norm inequality constraint (NICCB)
The Capon beamformer can experience significant performance degradation when there is a
mismatch between the presumed and actual characteristics of the source or array. The goal
of NICCB is to impose an additional inequality constraint on the Euclidean norm of w for


90
the purpose of improving the robustness to pointing errors and random perturbations in
sensor parameters, here w denotes the array weight vector. This requires incorporating a
norm inequality constraint on w of the form:

2
w (1.1)
where is the norm constraint parameter. Consequently, the NICCB problem is formulated
as follows:

2
min
. . 1
H
H
s t
w
w Rw
w s
w
(1.2)
where R is the data covariance matrix, s is the presumed signal steering vector, and ( )
H

denotes the conjugate transposition, denotes the vector
2
l norm. For the convenience of
analysis, and analyzing the choice of the norm constraint parameter, the solution to NICCB
[22] is introduced as follows.
1.2.1 Solution to NICCB
Let S be the set defined by the constraints in above optimization problem, namely:

{ }
2
1,
H
S = = w w s w (1.3)
Define a function:
( )
( ) ( )
2
1
, , 2
H H H
f = + + + w w Rw w w s s w (1.4)
where is the real-valued Lagrange multiplier, and 0 satisfying 0 + > R I so that
( )
1
, , f w can be minimized with respect to w . is the arbitrary Lagrange multiplier.
Then:
( )
1
, , ,
H
f S w w Rw w (1.5)
with equality on the boundary of S .
For the standard Capon beamformer

min
. . 1
H
H
s t
w
w Rw
w s
(1.6)
The optimal solution is:

1
1 H
=
R s
w
s R s
(1.7)


91
where
1
R is the inversion of R , namely ( )
1
denotes the matrix inversion. Here, we can
have:

( )
1 1 2
2
1 1 2
1
H
H
H
H H
H

= = =

R s R s s R s
w w w
s R s s R s
s R s
(1.8)
where
( )
2
2 1 1 1
= = R R R R , the above result using the Hermitian property of R .
Consider the condition:

( )
2
2
1
H
H

s R s
s R s
(1.9)
when the above condition is satisfied, the standard Capon beamformer solution (1.7)
satisfies the norm constraint of NICCB, hence, is also the solution to NICCB. For this case,
0 = and the norm constraint in NICCB is inactive.
Otherwise, we have the condition:

( )
2
2
1
H
H
<
s R s
s R s
(1.10)
which is an upper bound on so that NICCB is different from the standard Capon
beamformer. To deal with this case, we can rewrite the ( )
1
, , f w as follows:

( ) ( ) ( ) ( )
( )
1 1
1
1
2
, ,
2
H
H
f

= + + +

+ +
w w R I s R I w R I s
s R I s
(1.11)
Hence, the unconstrained minimizer of ( )
1
, , f w , for fixed and , is given by:
( )
1
,

= + w R I s (1.12)
Clearly, we have:
( ) ( ) ( )
1
2
2 1 ,
, , , 2 ,
H H
f f S
= = + + w s R I s w Rw w (1.13)
The maximization of ( )
2
, f with respect to . Hence, is given by:

( )
1
1

=
+ s R I s
(1.14)
Insert into ( )
2
, f , and let:

( ) ( )
( )
3 2
1
1
,
H
f f

= = +
+ s R I s
(1.15)


92
The maximization of the above function ( )
3
f with respect to gives:

( )
( )
2
2
1
H
H
+
=

+

s R I s
s R I s
(1.16)
Hence, the optimal Lagrange multiplier
can be obtained efficiently via, for example, a

Newtons method from the above equation of .
Note that using in
,

w yields:

( )
( )
1
1
+
=
+
R I s
w
s R I s
(1.17)
which satisfies the constraints of NICCB, namely:

1
H
= w s (1.18)
and

2
= w (1.19)
Hence,
w belongs to the boundary of S . Therefore,
w is our sought solution to the NICCB

optimization problem, which has the same form as the Capon beamformer with a diagonal
loading term I added to R, namely, NICCB also belongs to the class of diagonal loading
approaches.
From the above analysis, we can see that if the Lagrange multiplier is obtained, the
optimal weight vector for NICCB will be solved. In order to obtain the Lagrange multiplier
, we must solve the following equation via Newtons method, and let:

( )
( )
( )
2
2
1
H
H
h

+
=

+

s R I s
s R I s
(1.20)
Hence, the key problem of NICCB is finding the optimal Lagrange multiplier by above
equation (1.20). In this chapter, we will give the complete investigation on NICCB, and the
existence of its solution is analyzed as follows.
1.2.2 Solution to the optimal Lagrange multiplier
In order to solve the equation (1.20), we perform an eigenvalue decomposition (EVD) of the
sample covariance matrix as follows:

1
M
H H
i i i
i

=
= =
R U U u u
(1.21)
where ( )
1 2
, , ,
M
diag = " is diagonal matrix, ( )
1 2
, , ,
M
= U u u u " is an Hermitian
matrix,
i
( 1, 2 i M = " ) and
i
u ( 1, 2 i M = " ) are the eigenvalues and eigenvectors of R,


93
respectively, M is the total number of degrees-of-freedom. For the convenience of analysis,
we assume that the eigenvalues / eigenvectors of R are sorted in descending order, i.e. ,

1 2 M
" (1.22)
We can have:
( )
( ) ( )
2
2 2
1 1
2 2
2
1
1
H
H H M M
i
i i
i i
i i
H H M
H
M i i
i
i i
i i
h

= =
=
=
+ +
= =

+

+

s u
s u u s
s u u s
s u
(1.23)
Therefore, ( ) h is monotonically increasing function of 0 [22] , then:
( )
( ) ( )
2 2
2 2 2
1 1
1
2 2
2 2 2
1
1 1 1
1
H H
M M
i i
i i
i M
M
H M H H
M M
i i i
i
i i i
h

= =
=
= =
+ + +
= =

+

+ +

s u s u
s u s u s u
(1.24)
and
( )
( ) ( )
2 2
2 2 2
1 1
1
2 2
2 2 2
1
1
1 1
1
H H
M M
i i
i i
i M
M
H H H
M M
i i i
i
i M i i
h

= =
=
= =
+ + +
= =

+

+ +

s u s u
s u s u s u
(1.25)
and let:

2
1
M
H
i
i
=
=
s u (1.26)
Alternately, the above inequality relationship can be expressed as:

1
M

+
(1.27)
and

1
M

+
(1.28)


94
Next, we analyze the bound of the Lagrange multiplier and its existence.
1. If 1 > , using (1.27) and (1.28), we can have:

( )
( )
1
1
1 1
1
1
M
M
M M

+ +

+ +

(1.29)
Since 1 > , but
1
0
M
< , and 0 , so that
1 1
0
M M
> < .
Therefore, the bound of the Lagrange multiplier under
1
1
M
< < is given as
follows:

( ) ( ) 1 1
1
max min
0
1
M

= =
(1.30)
Then, we have:

( )
( )
( )
2
1
min
2
1
0
H
H
h h
= = >

s R s
s R s
(1.31)
and

( )
( )
( )
( )
( )
1
max
1
max
2
1
1
max
1
M
h h

+
= =

+

(1.32)
Hence, when
1
1
M
< < , there is unique solution
( ) ( ) 1 1
max min
,

satisfies ( ) h = .
2. If 1 < , using (1.27) and (1.28), we can have:

( )
( )
1
1
1 1
1
1
M
M
M M

+ +

+ +

(1.33)
Since 1 < , but
1
0
M
< , and 0 , so that
1 1
0
M M
> > .
Therefore, the bound of the Lagrange multiplier under
1
1
M
< < is given as
follows:

( ) ( ) 2 2
1
max min
0
1
M

= =
(1.34)
Then, we have:


95

( )
( )
( )
2
2
min
2
1
0
H
H
h h
= = >

s R s
s R s
(1.35)
and

( )
( )
( )
( )
( )
2
max
2
max
2
2
max
1
1
M
h h

+
= =

+

(1.36)
Hence, when
1
1
M
< < , there isnt a solution
(2) (2)
max min
,

satisfies ( ) h = .
In a word, we can conclude that when
1
1
M
< < , there is a unique solution
(1) (1)
max min
,

satisfies ( ) h = .
1.3 Norm inequality constraint parameter selection
From above analysis, we can see that it is important to select the norm inequality constraint
parameter for NICCB. If the norm inequality constraint parameter is large, it is
inactive. On the contrary, if the norm inequality constraint parameter is small, there isnt
a solution to satisfy NICCB.
We have analyzed that when
1
1
M
< < , there is a unique solution
(1) (1)
max min
,

satisfies ( ) h = . Hence, we can have the selecting bound of the norm inequality constraint
parameter as follows:

1
1
M
< <
(1.37)
Namely:

2
1
1 1
M

< <

(1.38)
Add the condition of
( )
2
0
2
1
H
H
< =
s R s
s R s
, we can obtain:

2
1
min 0 max
1 1
min ,
M

= < < =

(1.39)
If the norm inequality constraint parameter is out of the above bound, there is no solution
to NICCB. Hence, the norm inequality constraint parameter should be chosen in the
interval defined by the above inequalities.
1.4 Capon beamformer under norm equality constraint (NECCB)
From above analyses, we can see that the norm inequality constraint can enhance the
robustness of NICCB. Since the inequality relationship has a wide range, the norm of the


96
weight vector will vary in the relevant wide range. If the fluctuation of weight vector norm
is acutely, the performance improvement will be weakened greatly. Because the norm
equality constraint (NEC) is stronger than the norm inequality constraint (QIC), NECCB will
have more ascendant robust performance than NICCB. Hence, NECCB is proposed and is
solved effectively in this chapter.
NECCB is to impose an additional equality constraint on the Euclidean norm of w . The
NECCB problem is formulated as follows:

2
min
. . 1
H
H
s t
w
w Rw
w s
w
(1.40)
Compare NECCB with NICCB, we can educe the conclusion as follows: (1) The solution to
NICCB is obtained on the boundary of its constraint, similarly, for NECCB, the solution is
also obtained on its constraint boundary. (2) The solving methods of the two beamformers
(or the optimization problem) is different, such as the forenamed solution to NICCB, the
Lagrange multiplier of NICCB is taken as positive real-value only, but for NECCB, the
Lagrange multiplier is taken as arbitrary real-value, namely, it will be not only the positive
real-value, but also the negative real-value. Hence, if we analyze from the point of view of
the solving optimization problem, NECCB has two solutions to the optimal Lagrange
multiplier, one is positive, and another is negative. Actually, the positive one is the solution
to NICCB. For the sake of distinguishing the otherness, the negative solution is interested to
NECCB. In order to solve NECCB, we must make use of the discussed results of NICCB,
since the manipulation of some inequality, such as the inequality lessening and enlarging is
only right for the positive real-value when we solve NICCB.
Similar to NICCB, the solution to NECCB can also be solved by the Lagrange multiplier
methodology. And the optimal weight vector of NECCB has the same form as NICCB. The
difference between NECCB and NICCB is only the Lagrange multiplier
, for NICCB,
0 , here
is arbitrary real-value.
Although the solution to NECCB has the same form as NICCB, but the bound of the
Lagrange multiplier is different. In order to use the analyzed results of NICCB for NECCB,
replace the Lagrange multiplier by its absolute value, namely the bound of the Lagrange
multiplier
for NECCB is given by:

1
M

(1.41)
and

1
M

(1.42)
1. If 1 > , then:


97

1 1
1 1
M M

(1.43)
If
1
0
M
> , then
1 M
< , since
1
0
M
< , but 0 >
. Therefore, if
1
1
M
< < , we can have:

( ) ( ) 1 1
1 1
max min
1 1
M M

= =

(1.44)
Since
(1)
max
0 >
, and
(1) (1)
max min
0 = <

. Hence, when
1
1
M
< < , the solution to NECCB
in the bound of
( ) 1
max
0,

is the same as NICCB, but the solution in the bound of

( ) 1
min
, 0

is the true solution to NECCB.

2. If 1 < , then:

1 1
1 1
M M

(1.45)
If
1
0
M
> , then
1 M
> , since
1
0
M
< , but 0 >
. Therefore, if
1
1
M
< < , we can have:

( ) ( ) 2 2
1 1
max min
1 1
M M

= =

(1.46)
Since
( ) 2
max
0 >
, and
( ) ( ) 2 2
max min
0 = <

, with the above analysis of NICCB, we can obtain that
when
1
1
M
< < there isnt a solution in the bound of
( ) 2
max
0,

to NECCB, but the

solution in the bound of
( ) 2
min
, 0

is the true solution to NECCB.

From above analysis, we can conclude as follows:
1. When
1
1
M
< < , the solution in the bound of
( ) 1
min
, 0

is the true solution to

NECCN, and the norm equality constraint parameter should be chosen in the
interval defined by ( )
{ }
2
1 0
1 min ,
M
< < .
2. When
1
1
M
< < , the solution in the bound of
( ) 2
min
, 0

is the true solution to

NECCB, and the norm equality constraint parameter should be chosen in the bound
of ( ) { }
2
1 0
min 1 ,
M
< < .
3. NECCB has the form of diagonal loading with negative loading level, but NICCB has
the form of diagonal loading with positive loading level.


98
1.5 Simulation analysis
In order to validate the correctness and the efficiency of the proposed algorithms, we
analyze as follows. In our simulations, we assume a uniform linear array with N=10
omnidirectional sensors spaced half a wavelength apart. Through all examples, we assume
that there is one desired source, namely, there is a signal from direction 0, the Signal Noise
Ratio (SNR) is -5dB. Therein, the presumed signal direction is equal to 5 (i.e., there is a 5
direction mismatch).
For the comparison, the benchmark standard Capon beamforming algorithm that
corresponds to the ideal case when the covariance matrix is estimated by the maximun
likelihood estimator (MLE) and the actual steering vector is used, this algorithm does not
correspond to any real situation but is included in our simulations for the sake of
comparison only, and is denoted by Ideal-SCB in the figure. The other algorithms include:
standard Capon beamformer (SCB), NICCB, NECCB. For NICCB and NECCB, the constraint
parameter is selected as the median of the allowable bound.
1.5.1 Effectivity analyzing
In order to show the effectivity of the proposed algorithms, we first compare the pattern of
the mentioned Capon beamforming algorithms. The Capon beamformer pattern is given in
Fig. 1. Since the signal direction mismatch, the mainlobe of SCB departs from the signal
direction. The performance of NICCB is slightly better than SCB, and NECCB is the best of
all, the direction mismatch is overcame commendably and NECCB also has lower sidelobe
level. Here, NICCB uses the positive optimal loading level, NECCB uses the negative
optimal loading level. From the comparison, we can see that NECCB has the better
performance than NICCB.
0
-10
-20
-30
-40
-50
-60
B
e
a
m
f
o
r
m
e
r

p
a
t
t
e
r
n

(
d
B
)
-100 -50 0 50 100
Azimuth Angle (deg)
Ideal-SCB
SCB
NICCB
NECCB

Fig. 1. Capon beamformer pattern comparison


99
The variation of the beamformer SNR versus samples number is given in Fig. 2. We can see
that with the change of the samples number, the SNRs varies accordingly. The SNR of
NICCB is almost closed to the SNR of SCB, and is lower than the SNR of Ideal-SCB, but
NECCB is the best of all, especially for the small number, it has preferable performance.
Hence, the norm constraint can improve the SNR, and NECCB has the highest SNR among
the listed algorithms.
15
10
5
0
-5
-10
B
e
a
m
f
o
m
e
r

S
N
R

(
d
B
)
0 200 400 600 800 1000
Number of Samples
SNR: Ideal-SCB
SNR: SCB
SNR: NICCB
SNR: NECCB

Fig. 2. Output SNR versus samples number
Ideal-SCB
SCB
NICCB
NECCB
-10 -5 0 5 10
Angle Error (deg)
B
e
a
m
f
o
r
m
e
r

O
u
t
p
u
t

S
N
R

(
d
B
)
10
5
0
-5
-10
-15
-20
-25

Fig. 3. Output SNR versus angle mismatch


100
The variation of the Capon beamformer output signal-to-noise-ratio (SNR) versus signal
direction mismatch is given in Fig. 3. We can see that with the change of the signal direction
mismatch, the SNR varies accordingly, when the angle error is in the range of [-7, 7],
NECCB will has higher SNR than SCB, NICCB. The NECCB has the higher SNR can be
explained by the Fig. 1 of the beam pattern comparison, NECCB not only has the good
pointing performance, but also has the lower sidelobe level. Namely, for the same desired
signal output, the output noise of NECCB is lower. The simulation results can also be
explained as follows, for the used scene, the Signal Noise Ratio is -5dB, and for NECCB, the
optimal Lagrange is negative, namely the optimal loading level is negative, but for others,
the loading level is zero or positive. Therefore, for the NECCB beamformer, the output noise
power is decreased, but for other beamformers, the output noise power is increased. Hence,
the NECCB has higher output SNR than others. For the sake of saving space, the
corresponding beam pattern comparison isnt given, but in the simulation, NECCB pattern
also points to the actual signal direction exactly. Hence, NECCB has the better robustness in
the signal direction mismatch case.
From above analysis, we can see that NECCB has the best robustness against the signal
direction mismatch.
1.5.2 Correctness analyzing
Since NICCB and NECCB have the same form as that of SCB with diagonal loading. But
their key problems are how to find their own optimal loading level or Lagrange multiplier,
In order to show the impact of loading level on the Capon beamformer under norm
constraint (NCCB) and attest the correctness of the proposed algorithms, the simulation
results are given as follows.
The variation of the output SNR versus diagonal loading level is given in Fig. 4. We can see
that with the change of the loading level in the bound of
( ) ( ) 1 1
max min
,

, the SNR of NCCB
varies accordingly. When the loading level is positive, NCCB is NICCB, whereas, when the
loading level is negative, NCCB is NECCB. By comparison, we can see that NECCB has
higher SNR than NICCB, but for the optimal loading, namely when the loading level is
equal to -6.09, NECCB has the best pointing performance, and its SNR is the highest one,
where the optimal loading level -6.09 is calculated using the equation ( ) h = with
1
1
M
< < in the bound of
( ) 1
min
, 0

. Hence, the loading level has a great impact on

the SNR of the Capon beamformer, and determines the performance improvement.
The variation of the weight vector norm versus diagonal loading level is given in Fig. 5. We
can see that with the change of the loading level in the bound of
( ) ( ) 1 1
max min
,

, the weight
vector norm of NCCB varies accordingly. For most loading levels, the weight vector norm
varies slightly, but when the loading level is small in negative domain, the weight vector
norm is a little high, and the highest point is corresponding to the lowest point in Fig. 4.
Therefore, the loading level has a great impact on the weight vector norm, especially for the
negative loading.


101
-40 -30 -20 -10 0 10 20 30 40
Digital Loading Level ( )
B
e
a
m
f
o
r
m
e
r

S
N
R

(
d
B
)
6
4
2
0
-2
-4
-6
-8
Ideal-SCB
SCB
NCCB
Optimal
-6.09

Fig. 4. Output SNR versus loading level

-40 -30 -20 -10 0 10 20 30 40
Digital Loading Level ( )
W
e
i
g
h
t

V
e
c
t
o
r

N
o
r
m
1
0.9
0.8
0.7
0.6
0.5
0.4
0.3
Ideal-SCB
SCB
NCCB
Optimal
-6.09

Fig. 5. Weight vector norm versus loading level
From the above simulation results, we can see that the loading level has a great impact on
the performance of the Capon beamformer, and NECCB has the best pointing performance,
namely, the optimal negative loading is the best. This is also consistent to the theory
analysis, for the robust beamformer with diagonal loading, the improvement is determined
by the optimal loading level, when the loading level is optimal, the performance


102
improvement will be the optimal, but for other values, the improvement will be little, or
even worse.
1.5.3 Constraint parameter selection analyzing
For NCCB, there are two key problems, one is how to find the optimal loading level, and the
other is how to select the norm constraint parameter. Although we have solved the two
problems in theory, but there is another key problem, namely, how to select the optimal
norm constraint parameter. Therefore, the impact of norm constraint parameter on NCCB is
analyzed here particularly.
The variation of the output SNR versus norm constraint parameter is given in Fig. 6. We can
see that with the change of the norm constraint parameter in the allowable bound of
( )
min max
, , the SNR of the Capon beamformer varies accordingly. NICCB has a little
higher SNR than that of SCB, NECCB has the highest SNR. And with the norm constraint
parameter increasing, the SNR of NECCB increases correspondingly, but the SNR of NICCB
is inclined to the SNR of SCB. When the norm constraint parameter is equal to the
maximum, the constraint is inactive, and the three SNRs tend to the same value. Hence, the
SNR is determined by the choice of the norm constraint parameter, especially for NECCB.
B
e
a
m
f
o
r
m
e
r

S
N
R

(
d
B
)
6
4
2
0
-2
-4
-6
-8
Ideal-SCB
SCB
NICCB
NECCB
0.1 0.105 0.11 0.115 0.12 0.125 0.13 0.135 0.14
Norm Constrain Parameter ()

Fig. 6. Output SNR versus constraint parameter
The variation of the weight vector norm versus norm constraint parameter is given in Fig. 7.
When the norm constraint parameter is selected in the allowable bound of ( )
min max
, ,
the weight vector norms of NICCB and NECCB vary adaptively, and are equal to the square
root of the constraint parameter approximatively, this is consistent with the theory, namely
the solution is obtained on the constraint boundary. The slight difference is caused by the
approximative computation.


103
W
e
i
g
h
t

V
e
c
t
o
r

N
o
r
m
0.38
0.37
0.36
0.35
0.34
0.33
0.32
0.31
Ideal-SCB
SCB
NICCB
NECCB
Norm Contrain Value ()
0.1 0.105 0.11 0.115 0.12 0.125 0.13 0.135 0.14
Norm Constrain Parameter ()

Fig. 7. Weight norm versus constraint parameter
From above simulation results, we can see that if the norm constraint parameter is selected
in the allowable bound, the norm constraint parameter has a great impact on the
performance of NICCB and NECCB, especially for NECCB. But NECCB with the larger
constraint parameter has the better pointing performance, namely, when the constraint
parameter is selected as more larger in its allowable bound, the optimal negative loading
has the optimal improvement.
1.6 Conclusion
From the above analysis, we can conclude as follows. (I) The proposed algorithm is correct
and effective. (II) The norm constraint can improve the robustness of the Capon
beamformer. Especially, the equality constraint has the preferable improvement to
overcome the steering vector mismatch, and also has good robustness for the samples
number. (III) When the norm constraint parameter is selected in the allowable bound,
NECCB has the best performance, namely the optimal negative loading has the optimal
improvement, this is because that the norm equality constraint is stronger than the norm
inequality constraint.
2. Improved pattern synthesis method with linearly constraint minimum
variance criterion
Antenna pattern synthesis becomes the fundamental research contents with the wide
application of the array antenna in communication, radar and other areas, and catches the
attentions widely. The array antenna pattern synthesis is the task which solves the weight
values of the every element to force the antenna pattern inclining to the anticipant shape.
Dolph has first given the method of getting the weight function for uniform linear array to


104
achieve the Chebychev pattern

[39], therefore the optimal solution can be achieved in the
sense of giving the mainlobe width and the maximum lowest sidelobe level. However, how
to implement the pattern synthesis for the arbitrary array antenna efficiently is a challenging
research task in array signal processing society.
Currently, the methods of pattern synthesis can be classified as the two types, one is the
traditional vector weight methods

[40-42], the other one is the matrix weight methods [43],
therein, the intelligent computer methods are used to improve the calculating efficiency of
the optimal weight vector, such as the genetic evolution algorithms [44] and the particle
swarm optimization algorithms [45]. However, for any pattern synthesis method, the
iterative operation cant be avoided, and the iterative number determines the operation load
directly, the operation load, or titled as the compute efficiency is the key metric to evaluate
the validity of the pattern synthesis.
Guo Q et al propose the pattern synthesis method for the arbitrary array antenna with the
linearly constraint minimun variance criterion (LCMV-PS) [45], compared with the
traditional vector weight methods, this algorithm has the small iterative number and the
preferable convergence. However, by analysis and simulation, it is found that the iterative
coefficient determines its performance, namely, the iterative coefficient not only determines
the pattern shape, but also determines the iterative number, or titled as the compute load.
Therefore, how to select the iterative formula and its iterative coefficient is the key problem
to reduce the compute load and enhanced the applicability.
In this chapter, for the LCMV-PS method proposed in [45], by analyzing its implementation
and jammer power iterative formula, the improved fast robust LCMV-PS method is
proposed
[46]
. This algorithm takes into account the effect of the relative amplitude between
synthesis pattern and its reference upon the pattern synthesis adequately, via adding a
proportion constant to the iterative formula, the effect of their relative amplitude upon the
changing ratio of the jammer power is strengthened, not only the iterative efficiency of the
jammer power is improved, namely the iterative number is reduced, and the pattern
synthesis efficiency is improved, but also the selecting bound of the iterative coefficient is
extended, namely the effect of the iterative coefficient upon the pattern synthesis is
weakened, and the application area and applicability of the pattern synthesis method is
enhanced greatly. The last simulation attests its correctness and effectiveness.
2.1 Pattern synthesis method with LCMV criterion
The problem of array pattern synthesis can be simplified as follows, namely for the given
element number M and element position { }
1
M
i
i =
x , solving the complex weight vector w, and
force the array pattern ( ) P with the definite width and maximum value in the desired
direction, at the same time, make the sidelobe level according to the requirement.
The target of the pattern synthesis method for arbitrary arrays based on LCMV criterion
(LCMV-PS) is making all the sidelobe peak level equal to the minimum that the array can
achieve as possible. Furthermore, this method constructs many illusive jammers in the
sidelobe region, and the jammer power will be justed by the synthesis pattern amplitude in
its relative direction, namely, if the synthesis pattern amplitude is high in this direction, the
illusive jammer power will be increased. Therefore, the LCMV-PS method can be simple
described as follows:


105
1. Specify the mainlobe region [ ]
1 2
,
ML ML
and sidelobe envelope ( )
i
D .
Set the initial value of jammer power ( )
0 i
f , if
i
is in the sidelobe region, ( )
0
1
i
f = ,
otherwise if in the mainlobe region, ( )
0
0
i
f = , 1, i N = " , where N is the number of the
uniformly distributed jammers with one degree spacing, namely,
1 2
, , ,
N
" are the
degree values with the one degree spacing in the array pattern overlay region. And
( )
i
D is the given reference sidelobe envelope of the synthesis pattern.
2. Calculate jammer powers for the k-th iteration ( ) ( ) ( )
1 2
, , ,
k k k N
f f f " .
If k=0, then the jammer powers are the initial values ( ) ( ) ( )
0 1 0 2 0
, , ,
N
f f f " .
If k1, there is the iterative formula as follows:
( )
[ ]
( ) ( )
( )
[ ]
1 2
1
1 1 1 2
0 ,
max , 0 ,
ML ML
k k k 1
k k ML ML
k 1
f P Pr
f Kf
Pr

=

+

(2.1)
where ( )
1 k
f
is the jammer powers of the k-1-th iteration, K is the iterative coefficient.

( ) ( )
1
H
k
P
= w a is the pattern of the k-1-th iteration, therein w is the relative weight

vector, ( ) a is the steering vector, and ( )
H
denotes the conjugate transposition.
k 1
Pr

is the
sidelobe reference amplitude, if the arbitrary sidelobe shape is required in the pattern
synthesis, it is only to substitute ( ) ( )
k 1 k 1
Pr Pr D

= for
k 1
Pr

in the above formula, and
( ) D is the given sidelobe envelope.
3. Calculate the data covariance matrix
x
R , namely:
( ) ( ) ( )
1 2
, , ,
H
x k k k N
diag f f f = +

R A A I " (2.2)
where ( ) ( ) ( )
1 2
, , ,
N
=

A a a a " is the array manifold matrix. is a given small
quantity, and I is the identity matrix, I is added to prevent the covariance matrix from
being ill-conditioned.
4. Calculate the weight vector w according with the following LCMV beamforming
algorithm, then synthesize the pattern. If it is satisfactory, stop; otherwise, go to step (2)
and continue. Therein, w is solved by the below LCMV optimization problem, namely:

min
. .
H
H
s t
x
w
w R w
C w f
(2.3)
where C is the Mm constraint matrix, and f is the m1 constraint value vector. Its optimal
solution is:

( )
1
1 1 H

=
x x
w R C C R C f (2.4)
In the constraint condition of the optimization problem, the constraint of the mainlobe can
be imposed, the constraint of the sidelobe can also be added, in other words, the constraint
condition and parameter can be selected according to the pattern synthesis requirement.


106
2.2 Improvement of the jammer power iteration formula
From the step of the LCMV-PS method, we can see that the key is the jammer power
iteration in step (2), since it not only determines the synthesis pattern shape, but also
determines the final iterative number.
By the particular analysis of the LCMV-PS implementing steps, it is not difficulty to find
that although the relative difference of the synthesis pattern and the reference pattern
( ) ( )
k 1 k 1 k 1
P Pr Pr

is used as the ratio factor for the gain change, and to control the
change direction and quality of the jammer powers, actually, the expression of the iterative
formula ( ) ( ) ( ) ( ) ( )
1 1 1 k k k k 1 k 1
f Kf P Pr Pr

+ can be transformed as:
( ) ( )
( )
( )
( )
( )
1 1
1 1 1
1
k k 1 k
k k k
k 1 k 1
P Pr P
f Kf f K K
Pr Pr

+ = +

(2.5)
This expression indicates that the jammer powers between the adjacent iterations are
different by a proportional factor, when the iterative coefficient K is given, the jammer
power ratio of the adjacent iterations is determined by the relative amplitude of the
synthesis pattern and the reference pattern ( )
k 1 k 1
P Pr

. Therefore, for the given K, the
change of the jammer powers in the iteration process is determined by ( )
k 1 k 1
P Pr

, and
the relationship is a linear function.
From the pattern synthesis process of the LCMV-PS method, when the synthesis pattern is
higher than the reference pattern, the jammer powers should increase, and is in direct
proportion to the difference of the two patterns. When the synthesis pattern is more higher
than the reference pattern, the jammer powers should increase more larger. But when the
synthesis pattern is close to the reference pattern, the change of the jammer power should be
small, namely the adjustment should be precise at this time. Although the original iterative
formula is consistent to the analyzing idea, and the change ratio of the iterative jammer
powers is ( ) ( ) ( ) 1
k 1 k 1
K P Pr K

+ , namely is in direct proportion to ( )
k 1 k 1
P Pr

.
Therefore, for the original method, K is the main parameter to determine the iterative effect
and efficiency, and by the simulation, it is found that the synthesis pattern will be good
when the parameter K is selected in a small region, such as the reference value K=0.1 in [42].
Actually, for the difference element number or array parameter, the optimal value of K will
vary correspondingly. Hence, for the original method, how to select the optimal parameter
K is the key matter, it not only determines the effect of the synthesis pattern, but also
determines the efficiency of the jammer power iteration, namely the iterative number.
Since in the iterative process, it is the factor ( ) ( ) ( ) 1
k 1 k 1
K P Pr K

+ determining the
change quantity and direction of the jammer powers iteration, for the given K, the second
item is constant, but the first item is the linear function of ( )
k 1 k 1
P Pr

, and its
proportional coefficient is K, namely the slope K determines the change quantity of the
jammer power with ( )
k 1 k 1
P Pr

. With the slope of the linear function increasing, namely
for the given parameter K
p
>1, the change ratio of ( ) ( ) ( ) 1
p k 1 k 1
K K P Pr K

+ with
( )
k 1 k 1
P Pr

will be larger, namely the efficiency of the jammer power iteration will be
improved. At the same time, for the given parameter K
p
, when the effect of the jammer
power iteration is better, the selection of K will be loosened, namely K can be selected in a
wider region. Therefore, if the constant factor K
p
(K
p
>1) can be added as this method, the


107
efficiency of the jammer power iteration will be improved, and the bound for selecting the
iterative coefficient K will be enlarged, namely the selection of K will be simplified greatly.
Hence, in order to improve the iterative efficiency of the LCMV-PS method and simplify the
selection of the iterative coefficient K, the iterative formula of the jammer power can be
improved as follows, namely
( )
[ ]
( ) ( )
( )
[ ]
1 2
1 1 1 2
0 ,
max , 0 ,
ML ML
p k 1 k 1 k
k k ML ML
k 1
K P Pr f
f K f
Pr

=

+

(2.6)
where K
p
is used to adjust the effect of the relative amplitude ( )
k 1 k 1
P Pr

of the synthesis
pattern and reference pattern upon the change ratio of the jammer power, namely is used to
adjust the iterative efficiency of the pattern synthesis method, and other parameters have
the same sense as forenamed. If K
p
>1, the iterative efficiency will be advanced, whereas the
iterative efficiency will be reduced. It is important that the effect of the iterative coefficient K
upon the pattern synthesis is reduced greatly by adding the parameter K
p
, therefore, the
selection of parameter K will be simplified greatly.
Compared with the LCMV-PS method proposed in [42], the iterative formula of the jammer
power is added by a constant K
p
to adjust the iterative efficiency in this chapter, if K
p
>>1, the
efficiency of the proposed method will be improved greatly, therefore, the iterative number
will be reduced, so that the operation load will be reduced greatly by the proposed method.
At the same time, the bound for selecting K will also be enlarged greatly, and the application
area and applicability of the pattern synthesis method is enhanced.
2.3 Simulations
Since the proposed method has the higher iterative efficiency and stronger applicability as
compared with the LCMV-PS method proposed in [42], the simulation keystone is to
compare the iterative efficiency of the two methods, and the effect of the iterative coefficient
upon the two methods. The simulations are as follows.
2.3.1 Efficiency analyzing
In order to validate the efficiency of the proposed method, the single beam and multi-beam
pattern synthesis examples of the uniform and non-uniform linear array are given
respectively, and the single beam pattern synthesis examples of uniform and non-uniform
planar array are also given respectively. For the convenience of comparison, the proposed
improved LCMV-PS method is denoted as I-LCMV-PS, the LCMV-PS in [42] is denoted as
LCMV-PS, the reference pattern is denoted as Reference.
The single beam synthesis pattern of the uniform linear array is given in Fig. 8. Therein, the
element number is 32, the elements inter-space is half wavelength (/2), the mainlobe is
point to 0, the mainlobe width is 22, the iterative parameter of jammer power K=0.5, and
K
p
=100. When the optimal weight vector is solved under the LCMV criterion, the mainlobe
direction constraint is used only. Therein the iterative number of I-LCMV-PS is 5, but the
iterative number of LCMV-PS is 20.


108
0
-10
-20
-30
-40
-50
-60
-70
-80
-100 -80 -60 -40 -20 0 20 40 60 80 100
Azimuth Angle (deg)
Reference
I-LCMV-PS
LCMV-PS
B
e
a
m
f
o
r
m
e
r

P
a
t
t
e
r
n

(
d
B
)

Fig. 8. Single beam synthesis pattern of uniform linear array
The single beam synthesis pattern of the non-uniform linear array is given in Fig. 9. Therein,
the element number is 32, the element space vector is (/2)[0.29595, 1.5655, 2.7845, 3.9334,
4.999, 5.9753, 6.8645, 7.6764, 8.428, 9.1413, 9.842, 10.557, 11.311, 12.127, 13.021, 14.002, 15.073,
16.226, 17.449, 18.72, 20.017, 21.312, 22.579, 23.795, 24.939, 26, 26.971, 27.856, 28.664, 29.413,
30.125, 30.825], the mainlobe is also point to 0, the mainlobe width is also 22, and K=0.5,
K
p
=100. When the optimal weight vector is solved under the LCMV criterion, the mainlobe
direction constraint is used only. The iterative number of the two methods are 5 and 20,
respectively.

0
-10
-20
-30
-40
-50
-60
-100 -80 -60 -40 -20 0 20 40 60 80 100
Azimuth Angle (deg)
Reference
I-LCMV-PS
LCMV-PS
B
e
a
m
f
o
r
m
e
r

P
a
t
t
e
r
n

(
d
B
)

Fig. 9. Single beam synthesis pattern of non-uniform linear array


109
In order to attest the applicability of the improved algorithm to the planar array, and its
effectivity of multi-beam synthesis, namely the arbitrary array and arbitrary pattern
synthesis, the particular simulation examples are given as Fig. 11.~Fig. 16.
The multi-beam synthesis patterns of the uniform and non-uniform linear array are given in
Fig. 11. and Fig. 12. The parameters are same as Fig. 8. and Fig. 9., the two beams point to
45 and -45 respectively. Therein the iterative number of I-LCMV-PS is 6, but the iterative
number of LCMV-PS is 25.
Y

D
i
r
e
c
t
i
o
n

(
C
o
l
u
m
n
)
0.14
0.12
0.10
.008
0.06
0.04
0.02
0
X Direction (Row)
0 0.02 0.04 0.06 0.08 0.10 0.12 0.14

Fig. 10. Element position of the non-uniform planar array
0
-10
-20
-30
-40
-50
B
e
a
m
f
o
r
m
e
r

P
a
t
t
e
r
n

(
d
B
)
-100 -50 0 50 100
Azimuth Angle (deg)
Reference
I-LCMV-PS
LCMV-PS

Fig. 11. Multi-beam synthesis pattern of uniform linear array


110
0
-5
-10
-15
-20
-25
-30
-35
B
e
a
m
f
o
r
m
e
r

P
a
t
t
e
r
n

(
d
B
)
-100 -50 0 50 100
Azimuth Angle (deg)
Reference
I-LCMV-PS
LCMV-PS

Fig. 12. Multi-beam synthesis pattern of non-uniform linear array
The single beam synthesis patterns of the uniform and non-uniform planar array with
I-LCMV-PS are given in Fig. 13.~ Fig. 16. Therein, they have the same element number 36,
the uniform planar array is phalanx, and element space is half wavelength, but the element
position of the non-uniform planar array is given as Fig. 10. The mainlobe of the two array
point to (0,0), and the beam-widths in the azimuth and elevation direction are all 30. In
the simulation, the iterative number of I-LCMV-PS is 8, and the Fig. 14. and Fig. 16. is the
side view figure of Fig. 13. and Fig. 15. respectively.

B
e
a
m
f
o
r
m
e
r

P
a
t
t
e
r
n

(
d
B
)
0
-20
-40
-60
100
0
-100
Elevation Angle (deg)
Azimuth Angle (deg)
-100
100
0
-50
50

Fig. 13. Single beam synthesis pattern of uniform planar array (1)


111

0
-10
-20
-30
-40
-50
-60
B
e
a
m
f
o
r
m
e
r

P
a
t
t
e
r
n

(
d
B
)
-100 -50 0 50 100

Fig. 14. Single beam synthesis pattern of uniform planar array (2)

B
e
a
m
f
o
r
m
e
r

P
a
t
t
e
r
n

(
d
B
) 0
-10
-20
-30
-40
Elevation Angle (deg) Azimuth Angle (deg)
100
0
-100
100
50
0
-50
-100

Fig. 15. Single beam synthesis pattern of non-uniform planar array (1)
From the above simulations, we can see that the two methods have the preferable synthesis
pattern, but I-LCMV-PS has small iterative number, namely it has the higher pattern
synthesis efficiency.


112

0
-5
-10
-15
-20
-25
-30
-35
B
e
a
m
f
o
r
m
e
r

P
a
t
t
e
r
n

(
d
B
)
-100 -50 0 50 100

Fig. 16. Single beam synthesis pattern of non-uniform planar array (2)
2.3.2 Convergence analyzing
In order to compare the convergence characteristic, limit by the chapter length, here the
example of the uniform linear array is given, about the examples of the non-uniform linear
array and the planar array are similar to the uniform linear array.
B
e
a
m
f
o
r
m
e
r

P
a
t
t
e
r
n

(
d
B
)
0
-10
-20
-30
-40
-50
-60
-70
-80
-90
-100 -80 -60 -40 -20 0 20 40 60 80 100
Azimuth angle (deg)
1
2
3
4

Fig. 17. Synthesis pattern versus iterative number of I-LCMV-PS


113
Fig. 10. and Fig. 11. give the convergence of the synthesis pattern for the uniform linear array
with I-LCMV-PS and LCMV-PS respectively. Therein, the parameters are as same as 2.3.1.
From Fig. 10., we can see that when the iterative number is larger than 4, I-LCMV-PS can
achieve the preferable pattern synthesis effect, but for LCMV-PS, the iterative number must
be larger than 20, because I-LCMV-PS has the higher efficiency of the jammer power iteration.
B
e
a
m
f
o
r
m
e
r

P
a
t
t
e
r
n

(
d
B
)
0
-10
-20
-30
-40
-50
-60
-70
-80
-90
-100 -80 -60 -40 -20 0 20 40 60 80 100
Azimuth angle (deg)
1
2
3
4

Fig. 18. Synthesis pattern versus iterative number of LCMV-PS

-10
-20
-30
-40
-50
-60
S
y
n
t
h
e
s
i
s

E
r
r
o
r

(
d
B
)
0 10 20 30 40 50 60 70 80
Iteration Number
I-LCMV-PS
LCMV-PS

Fig. 19. Pattern synthesis error versus iteration number


114
In order to attest the convergence and synthesis precision of the improved algorithm, Fig.
19. gives the pattern synthesis error versus the iterative number. Therein, the pattern
synthesis error is calculated as follows:
( ) ( )
1
N
sum k i i
i
E P Pr
=
=
(2.7)
where ( ) Pr is the reference pattern, ( )
k
P is the synthesis pattern after k-th iteration.
From the error curve, we can see that the improved algorithm has the fast convergence
performance, and fall rapidly from beginning, at last, the two curves converge at the same
value, namely the convergence is consistent with the Fig. 17. and Fig. 18. Therefore, the
proposed algorithm has the good convergence and synthesis precision.
In order to analyze the effect of the iterative coefficient upon the pattern synthesis, Fig. 20.
and Fig. 21. give the synthesis pattern versus iterative coefficient for the uniform linear array
with I-LCMV-PS and LCMV-PS respectively. Therein, the parameters are as same as 2.3.1.
From Fig. 20., we can see that when K
p
=100, the selection of iterative coefficient K has very
little effect upon the pattern synthesis, in simulation, when 0.005<K<2000, the preferable
performance can be achieved, the 2000 is the upper bound in simulation, if the parameter is
larger than this value, the good performance can also be achieved. Actually, if K
p
larger, the
selecting bound for K wll be wider, it is consistent with the theory analysis. But from Fig.
21., we can also see that the efficiency of the pattern synthesis is determined by K, in the
simulation, we find that when 0.1<K<1.6, the preferable effect is achieved. By the
comparison, we can see that the improved LCMV-PS method has the lower dependence
upon the iterative coefficient, and enables the selecting bound for the iterative coefficient K
enlarged greatly, and enhanced its application area and applicability.

B
e
a
m
f
o
r
m
e
r

P
a
t
t
e
r
n

(
d
B
)
0
-10
-20
-30
-40
-50
-60
-70
-80
-100 -80 -60 -40 -20 0 20 40 60 80 100
Azimuth angle (deg)
0.001
0.010
100.000
2000.000

Fig. 20. Synthesis pattern versus iterative coefficient of I-LCMV-PS


115

B
e
a
m
f
o
r
m
e
r

P
a
t
t
e
r
n

(
d
B
)
0
-10
-20
-30
-40
-50
-60
-70
-80
-100 -80 -60 -40 -20 0 20 40 60 80 100
Azimuth angle (deg)
0.001
0.100
1.000
2.000

Fig. 21. Synthesis pattern versus iterative coefficient of LCMV-PS
2.4 Conclusion
From the above analysis and simulation, we can conclude as follows: (I) The proposed
jammer power iterative formula is correct and effective. (II) By the improvement for the
iterative formula of the original method, the iterative efficiency is increased greatly, and the
iterative number is reduced greatly, therefore the operation load of the pattern synthesis is
reduced efficiently. (III)The improved jammer power iterative formula enlarges the selecting
bound for the iterative coefficient, and reduces the effect of the parameter upon the pattern
synthesis, and enhances the application area and applicability of the proposed pattern
synthesis method.
3. Unitary Root-MUSIC
The problem of estimating the direction-of-arrival (DOA) of narrowband sources from
sensor array data has received considerable attention. The eigen-based methods for DOA
estimation represent a class techniques that offer a much better resolution performance than
that of conventional beamformers. In eigen-based methods, signal and noise subspaces are
identified first via a M M generalized EVD (GEVD) of the array data/noise correlation
matrix pencil, where M equals the number of array elements. A search is then conducted
over a null spectrum associated with the noise subspace, to locate the minima, from which
the source DOAs can be determined. In the case where a uniform linear array (ULA) is
employed, the null-spectrum searching can be converted into a polynomial rooting problem.
Two well known examples are the Root-MUSIC
[47]
and Root-Minimum-Norm
[48]
method.
They belong to the so-called weighted root-form eigen-based methods. Compared to their
spectrum-searching or spectral-form counterparts, root-form methods exhibit a higher
resolution capability in dealing with closely spaced sources.Rao and Hari
[49]
argue that a


116
zero of the null spectrum, having a large radial error, will cause the corresponding spectral
minima to be less defined,but does not affect the DOA estimates. As for the mean-squared
errors of the DOA estimates,Li and Vaccaro
[50]
show that both spectral and root-form
methods yield the same expression. It should be borne in mind, however, that the result
holds only when each of the sources has a minimum corresponding to it in the null
spectrum.
A major issue regarding eigen-based methods is the heavy computational load associated
with the GEVD. This is more significant when M is large. To remedy this the concept of
beamspace transformation was proposed
[51]
as a means of reducing the dimension of the
array data. Ta.S.L proposed a novel iterative implementation of beamspace root-form
methods without the need for large-order polynomial rooting
[52]
. Marius.P, Alex.B.G and
Martin.H proposed the unitary root MUSIC algorithm reduces the computational
complexity because it exploits the eigendecomposition of a real-valued covariance matrix
[53]
.
In this chapter [54], combining the algorithms of Root-MUSIC and Unitary-root MUSIC, the
Root-MUSIC algorithm with real-valued eigendecomposition is given.
3.1 Array signal model
Assume a uniform linear array (ULA) is composed of M sensors, and let it receive q
( q M < ) narrowband signals impinging with unknown directions of arrival ((DOA)
1
,
2
, ,
q
. Assume that there are N snapshots ( ) 1 x , ( ) 2 x , , ( ) N x available. The tth
measured snapshot of the array is generally modeled as:
( ) ( ) ( ) t t t = + x As n (3.1)
where ( )
( )
1
, ,
q

=

A a a " is the M q composite steering matrix, the columns of
which represent a basis for the signal subspace, ( ) a represents the arrays 1 M complex
manifold:

( )
( ) ( ) ( ) 2 sin 2 1 sin
1, , ,
T
j d j d M
e e

=

a "
(3.2)
In addition
( ) t s denotes the 1 q vector of source waveforms;
( ) t n denotes the 1 M vector of white sensor noise;
is the wavelength;
d is the interelement space;
( )
T
denotes transpose.
It is generally assumed that signals are uncorrelated with the noise ( ) t n . The sensor noise is
assumed to be a zero-mean spatially and temporally uncorrelated Gaussian process with the
uunknown diagonal covariance matrix given by

( ) ( )
{ } { }
2 2 2 2
, , , ,
H
n
E t t diag = = = R n n I "
(3.3)
where {} E is the expectation operator,and ( )
H
stands for the Hermitian transpose, I is the
identity matrix,
2
is the noise variance.


117
Data model (1) allows us to write the covariance matrix of the array measurements as:

( ) ( )
{ }
2
H
H
E t t = = + R x x ASA I
(3.4)
where ( ) ( )
{ }
H
E t t = S s s is the q q source waveform covariance matrix.
3.2 Root-MUSIC
Root-MUSIC is the polynomial rooting form of MUSIC, namely, the spectrum peak
searching is replaced by polynomial rooting in MUSIC implementation.
In Root-MUSIC, the polynomial should be defined as follows
( ) ( ) 1, ,
H
i i
f z z i q M = = + u a " (3.5)
where,
i
u is the eigenvector corresponding to the M-q minimum eigen-value of the data
covariance matrix, and

( )
1
1 ,
T
M
z z z

=

a "
(3.6)
From the above definition, we can include that the polynomial roots lie on the unit circle
properly when exp( ) z j = , and ( )
j
e

a is the steering vector of space frequency . From the
eigen-space algorithms, ( )
m
j
m
e

= a a is the signal steering vector, and it is orthogonal to the
space of the noise. Therefore, the polynomial definition can be modified as the following
form
( ) ( ) ( )
H H
N N
f z z z = a U U a (3.7)
where
N
U is the noise space, namely, let the eigendecomposition of the matrix R be
defined in a standard way

2 H H H
S S S N N
= = + R U U U U U U (3.8)
where
1 S q

=

U u u " ,
1
,
N q M +

=

U u u " ,
{ }
1
diag , ,
S q
= "
and the subscripts S and N stand for signal- and noise-subspace, respectively.
Therefore, the source DOA information can be obtained when the above roots are solved. At
the same time, we found the item of z
in the polynomial, and the solving process will

become complex and difficult. In order to simplify problem, the above polynomial can be
modified as

( )
( )
( )
1 1 M T H
N N
f z z z z

= a U U a
(3.9)
Here, the polynomial order is 2(M-1), and has (M-1) pairs roots, the every pair roots have
the mutual conjugate relationship. In the (M-1) pairs roots, there are q roots
1
, ,
q
z z " are
distributed on the unit circle, and


118
( ) exp , 1
i i
z j i q = (3.10)
For the ULA, the corresponding DOA of signal can be calculated as
( ) arcsin arg , 1, ,
2
i i
z i q
d

= =

" (3.11)
where is the signal wavelength, d is the array space.
A simple alternative method is proposed in Ref.[55]. From above analysis, we can see that
the signal space is orthogonal to the noise space, therefore
exp 2 sin , 1, ,
i i
d
z j i q

= =

" (3.12)
should be q roots of all M-q polynomials in Eq. (3.9), namely

( )
0, 1, , , 1, ,
i j
f z j q i q M = = = + " " (3.13)
Eq.(3.9) represents M-q polynomials of M-1 order. From Eq.( 3-13), they should have a q-
order maximum common factor, which can be denoted as ( ) f z . The DOAs of all the
sources can be obtained by rooting ( ) f z . From the eigenvectors of the noise space, ( ) f z
can be obtained as follows
There exists a vector
1
T
M q
b b

=

b " which satisfies

1
1
1
2
1 0 0
H
T
N H
N q
H
M
N
c c

= =

U
U b b
U
" " (3.14)
where
1 N
U is q(M-q) sub-matrix and
2 N
U is (M-q)(M-q) sub-matrix of
N
U . This can be
understood by noticing that the product of
N
U and b represents a linear combination of
noise vectors represented in the M-q dimensional noise space. The product
2 N
U and b
defines a system of M-q equations with M-q unknowns. b can be fixed to be the solution
that results in a product
[ ] 1 0
T
" . The product of
1 N
U and b is then a set of coefficients
that are determined.
Adopting this approach b is obtained by

[ ]
( )
1
2
1
1 0 0
T
N
M q

= b U "
(3.15)
And
1 q
c c

=

c " is determined easily as

[ ]
( )
1
1 1 2
1
1 0 0
T
N N N
M q

= = c U b U U "
(3.16)
Now that c has been determined, the polynomial ( ) f z is formed by

( )
1
1
1
1
, 1
q
i
i M
i
f z c z c
+
+
=
= =
(3.17)


119
Eq.(3.17) has q roots which are correspond to the DOAs of q sources. After obtaining q roots
of Eq.(3.13), { }
1
q
i
i
z
=
, the DOAs of the sources are obtained by Eq.(3.12).
From the well known conventional Root-MUSIC polynomial of Eq.(3.7), we can conclude
that it is a function of z, namely

( ) ( ) ( ) ( )
{ }
( )
1 1
Root MUSIC
1 1 1
M T H M T H
N N S S
f z z z z z z z

= = a U U a a U U a
(3.18)
Here the orthogonal property of the signal and noise subspace is used.
3.3 Root-MUSIC with real-valued eigendecomposition
The matrix R is the centro-Hermitian if

= R JR J (3.19)
where J is the exchange matrix with ones on its antidiagonal and zeros elsewhere, and
( )
stands for complex conjugate. The matrix (3.3)is known to be centro-Hermitian if and
only if S is a diagonal matrix, i.e., when the signal source are uncorrelated. However, to
double the number of snapshots and decorrelate possibly correlated source pairs in the
case of an arbitrary matrix S , the centro-Hermitian property is sometimes forced by means
of the so-called forward-backward (FB) averaging:

( )
~
2
1
2
H
FB

= + = + R R JR J ASA I
(3.20)
where

( )
~
1
2

= + S S DS D
(3.21)

( ) ( ) ( ) ( )
{ }
1
2 1 sin 2 1 sin
, , ,
q
j d M j d M
diag e e

= D "
(3.22)
Let us define the matrix as:

=
H
FB
C Q R Q
(3.23)
therefore, the C is a real-valued covariance matrix, where Q is any unitary,column
conjugate symmetric M M matrix. Any matrix Q is column conjugate symmetric if

= JQ Q
(3.24)
For example, the following sparse matrices

2
1
2
n
j
j

=

I I
Q
J J
(3.25)

2 1
1
2
2
T T
n
j
j
+

=

I 0 I
Q 0 0
J 0 J
(3.26)


120
can be chosen for arrays with an even and odd number of sensors,respectively, where the
vector ( ) 0, 0, , 0
T
= 0 "
From (3.23), and insert (3.20) to it, it follows that

( ) ( )
1 1
2 2

= = + = +

H H H H
FB
C Q R Q Q R JR J Q Q RQ Q JR J Q (3.27)
using
= JQ Q ,
= Q JQ and
H
= J J , we obtain that

( ) ( )
( ) ( ) ( )
1 1
2 2
1 1
2 2
Re

= + = +

= + = +

=

H
H H H
H
H H H
H
C Q RQ Q JR JQ Q RQ JQ R JQ
Q RQ Q R Q Q RQ Q RQ
Q RQ
(3.28)
therefore, we prove that C is a real-valued covariance matrix.
Let the eigendecompositions of the matrices R,
FB
R and C be defined in a standard way

2 H H H
S S S N N
= = + R V V V V V V (3.29)

2 H H H
FB S S S N N
= = + R U U U U U U (3.30)

2 H H H
S S S N N
C = = + E E E E E E (3.31)
where
1
, ,
S q

=

V v v " ,
1
, ,
N q M +

=

V v v " ,
{ }
1
, ,
S q
diag = "
1
, ,
S q

=

U u u " ,
1
, ,
N q M +

=

U u u " ,
{ }
1
, ,
S q
diag = "
1
, ,
S q

=

E e e " ,
1
, ,
N q M +

=

E e e " ,
{ }
1
, ,
S q
diag = "
and the subscripts S and N stand for signal- and noise-subspace,respectively.
Assume the Characteristic equation for the matrix
FB
R as
=
FB
R u u (3.32)
we can obtain that

H H H
= =
FB
Q R u Q u Q u (3.33)
with the use of equation: =
H
QQ I and the definition of C,we obtain that

H H H
= = =
H H
FB FB
Q R u Q R QQ u C Q u Q u (3.34)
Equation
H
=
H
C Q u Q u can be identified as the characteristic one for the real-valued
covariance matrix C.


121
Hence, using (3.30), (3.31), (3.32) and (3.34), the eigenvectors and eigenvalues of the matrices
FB
R and C are related as

H
= E Q U (3.35)
= (3.36)
It is well known that the conventional Root-MUSIC polynomial is given by
( ) ( ) ( ) 1
T H
MUSIC N N
f z z z = a V V a (3.37)
( )
{ }
( ) 1 1
T H
S S
z z = a V V a (3.38)
where
( )
1
1, , ,
T
M
z z z

=

a " (3.39)
j
z e

= , and ( ) 2 sin d = . Similarly to(3.37) and (3.38), the FB root-MUSIC polynomial
can be used:
( ) ( ) ( )
1
1
M T H
FB MUSIC N N
f z z z z
= a U U a (3.40)
( )
{ }
( )
1
1 1
M T H
S S
z z z
= a U U a (3.41)
A simple manipulation with the use of (3.35) and =
H
QQ I , we can obtain that:
( ) ( ) ( )
1
1
M T H
FB MUSIC N N
f z z z z
=
H H
a QQ U U QQ a (3.42)
( )
( ) ( )
( )
1
1
M T
N N
z z z
=
H
H H H
a Q Q U Q U Q a (3.43)
( ) ( )
1
1
M T H
N N
z z z
=
H
a Q E E Q a (3.44)
( ) ( )
~ ~
1
1
T
M H
N N
z z z
= a E E a (3.45)
( )
C MUSIC
f z
= (3.46)
where the manifold
( ) ( )
~
z z =
H
a Q a (3.47)
should be exploited for the polynomial rooting in (3.45). The relationship between the
former and the new manifolds follows from the expression for the real-valued covariance
matrix (3.23). From (3.23)and (3.20), we have


122

~ ~
2 2
~ ~ ~
2
H H
H

= = + = +

= +
H H H H
FB
H
C Q R Q Q ASA I Q Q ASA Q Q Q
ASA Q Q

(3.48)
(3.49)

where

~
H
= A A Q (3.50)
Let us term the polynomial (3.46) as the polynomial of Root-MUSIC with real-valued
eigendecomposition (RVED-Root-MUSIC), since it exploits the eigendecomposition of the
real-valued matrix (3.24) instead of that of the complex matrices (3.18) or (3.20). But from
(3.42) to (3.44), it is clear that the FB and RVED-Root-MUSIC polynomials are identical.
Hence, the performance of RVED-ROOT-MUSIC does not depend on a particular choice of
the unitary column conjugate symmetric matrix Q.
3.4 Polynomial coefficient finding
From (3.44) and (3.45), we obtain the polynomial of RVED-Root-MUSIC, which is a function
of z . The next thing is finding the coefficient of the polynomial
[56]
..
Using (3.44), we have:

( ) ( ) ( )
( ) ( )
1
1
1
1
M T H
C MUSIC N N
M T
f z z z z
z z z
=
=
H
a Q E E Q a
a G a
(3.51)
where

( )
,
H
N N i j
M M
g
= =
H
G Q E E Q (3.52)
Inserting (3.39) into (3.52), and with simple manipulation, we obtain that
( )
( ) ( )
( )
( )
1 1 1 1 1
1,1 1,
1 1 2 1
,1 ,
1
,1 ,2 ,
1 1 1
1
, 1
1, , , 1, , ,
, , , 1 1, , ,
1
, , ,
1
T
M M M
C MUSIC
M
T
M M M
M M M
M M M
M i M i M i
i i i M
i i i
M
M
i
f z z z z z z
g g
z z z z
g g
z
g z g z g z
z
g z

= = =

=

=

=

=

G " "
"
" # " # "
"
"
#
( ) 1 1
,2 ,
1 1 1
M M M
M i M i M i
i i M
i i i
z g z z g z

= = =

+ + +

"


123
the polynomial of RVED-Root-MUSIC is given by
( ) ( )
( )
( )
( )
( )
( )
( )
( )
( )
2 2 0
1,
2 2 1
2, 2, 1
2 2 2
3, 2, 1 1, 2
2 2 1
, 1, 1 2,2 1,1
2 2
, 1 1, 2 3,2 2,1
2 2 1
, 1 1, 2 3,2 2,1
2
,2 1,1
M
C MUSIC M
M
M M
M
M M M
M M
M M M M
M M
M M M M
M M
M M M M
M
M M
f z g z
g g z
g g g z
g g g g z
g g g g z
g g g g z
g g z

=
+ +
+ + +
+
+ + + + +
+ + + + +
+ + + + +
+
+ +
"
"
"
"
"
( )
( )
( )
2 3
2 2 2
,1
M M
M M M
M
g z
+
+
+

So the number of coefficient of the polynomial is 2 1 M , and the computation of the
coefficient is given as follows

( )
,
1
2 1 1
,
1
1, 2, ,
1, 2, , 2 1
k
i M k i
i
k
M k
k M i i
i
g k L
a
g k L L L
+
=
+
+
=
= + +
"
"
(3.53)
where
k
a denotes the kth coefficient of the polynomial.
Based upon our analysis, using (3.4), (3.20), (3.23), (3.35), (3.26), (3.31), (3.35), (3.44), (3.51),
(3.52) and (3.53), the fast algorithm for RVED-Root-MUSIC can be formulated as the
following seven-step procedure:
Step 1. Compute R and
FB
R with the use of (3.4) and (3.6).and the estimate is given by
( ) ( )
1
1
N
H
k
k k
N
=
=

R x x then
1
2
FB

= +

R R JR J .
Step 2. Compute C, and the Q is dependent on the number of array sensors. The estimate
of the real-valued covariance matrix is given by

=
H
FB C Q R Q
Step 3. Obtain
N
E from the eigendecomposition of C.and the estimate of
N
E ,
E is given
by the eigendecomposition of
C
Step 4. Compute G with the use of (3.52). And the estimate of G,
G is given by
H
N N

=
H
G Q E E Q


124
Step 5. Compute the coefficient of the polynomial by (3.53).
Step 6. Find the root of the polynomial (3.51), and select the q roots that are nearest to the
unit circle as being the roots corresponding to the DOA estimates.
Step 7. DOA estimate, using:
( ) arcsin arg 1, ,
2
k k
z k q
d

= =

"
where
k
z represents one of the q roots selected for DOA estimation.
From the above analysis, we can conclude that the RVED-Root-MUSIC has a lower
computational complexity than the conventional root-MUSIC technique thanks to the
eigendecomposition of the real-valued matrix instead of that of the complex matrices, and
the asymptotic performance of it is better than of conventional root-MUSIC due to the FB
averaging effect..
3.5 Simulations
In this section, we present some simulation results to illustrate the performance of RVED-
Root-MUSIC. We consider a ULA with M=8 elements and the inter-element space is equal
to a half of wavelength. There are three signals with SNRs of 30 dB impinges on the array
from
1
80 = ,
2
20 = ,
3
40 = . The detailed simulation results are shown as Fig. 22. ~
Fig. 25.
0.35
0.30
0.25
0.20
0.15
0.10
0.05
0
D
O
A

D
e
p
a
r
t
u
r
e
RVED Root MUSIC
-80
-20
40
0 200 400 600 800 1000
Snapshot Number

Fig. 22. DOA departure vs dnapshot number. Signal DOA=[-80 -20 40], SNR=5dB


125
0.35
0.30
0.25
0.20
0.15
0.10
0.05
0
D
O
A

D
e
p
a
r
t
u
r
e
Root MUSIC
-80
-20
40
0 200 400 600 800 1000
Snapshot Number

Fig. 23. DOA departure vs snapshot number. Signal DOA=[-80 -20 40], SNR=5dB
0.5
0.4
0.3
0.2
0.1
0
D
O
A

D
e
p
a
r
t
u
r
e
RVED Root MUSIC
-80
-20
40
-10 -5 0 5 10
SNR

Fig. 24. DOA departure vs SNR. Signal DOA=[-80 -20 40], Snapshot number =1000


126
0.5
0.4
0.3
0.2
0.1
0
D
O
A

D
e
p
a
r
t
u
r
e
Root MUSIC
-80
-20
40
-10 -5 0 5 10
SNR

Fig. 25. DOA departure vs SNR. Signal DOA=[-80 -20 40], Snapshot number =1000
Fig. 22. and Fig. 23. depict DOA departure versus snapshot number results of
RVED-Root-MUSIC and Root-MUSIC respectively, where the SNR=5dB. In figure 22. and
23., the x-axis denotes the snapshot number, and y-axis denotes the departure of signal
DOA.
Fig. 24. and Fig. 25. depict DOA departure versus SNR results of RVED-Root-MUSIC and
Root-MUSIC respectively, where the snapshot number =1000. In figure 24. and 25., the
x-axis denotes the SNR, and y-axis denotes the departure of signal DOA .
From the detecting results and the comparison between RVED-Root-MUSIC and Root-
MUSIC, we can conclude that RVED-Root-MUSIC can detect DOA of signal quickly and
effectively. At the same time, the results validate the correctness and effective of this
algorithm.
3.6 Conclusion
An improved version of the Root-MUSIC algorithm, called Root-MUSIC with real-valued
eigendecomposition (RVED-Root-MUSIC), has been presented in this chapter. The
computational complexity is reduced significantly by exploiting the one-to-one
correspondence between centro-Hermitian and real matrices, allowing a transformation to
real matrices, which can be maintained for all steps of the algorithm. Due to the inherent
forward-backward averaging effect, RVED-Root-MUSIC can separate two completely
coherent sources and provide improved estimates for correlated signals.


127
4. Real-value space ESPRIT algorithm and its implement
The recovery of signal parameters from noisy observations is a fundamental problem in
(real-time) array signal processing. Due to their simplicity and high-resolution capability,the
subspace estimation schemes have been attracting considerable attention. Among them the
most representative are MUSIC and ESPRIT methods. MUSIC utilizes the orthogonal
characteristic of noisy subspace of data covariance matrix,but ESPRIT exploits the rotational
invariance structure of the signal subspace
[57,58]
. The virtue of ESPRIT is the low
computational burden,and not requiring spectrum peak searching by contrast with MUSIC.
Comparing with Root-MUSIC, ESPRIT obtains the information of signal direction of
arriving (DOA) via exploiting the rotational invariance of every subarray (every subarray s
signal subspace), but Root-MUSIC estimates the signal DOA by solving the polynomial,
which is constructed by using the orthogonal between the steering vector and noise
subspace.
Unitary ESPRIT achieves even more accurate results than previous ESPRIT techniques by
taking advantage of the unit magnitude property of the phase factors that represent the
phase delays between the two subarrays
[59]
. It has been shown in [63] that constraining the
phase factors to the unit circle can also give some improvement for correlated sources. For
centro-symmetric sensor arrays with a translational invariance structure, Unitary ESPRIT
provides a very simple and efficient solution to this task.
Although Unitary ESPRIT effectively doubles the number of data samples, the computational
complexity is reduced by transforming the required rank-revealing factorizations of
complex matrices into decompositions of real-valued matrices of the same size. Thus, we
obtain increased estimation accuracy with a reduced computational load. This reduction can
be achieved by constructing invertible transformations that map centro-Hermitian matrices
to real matrices.
The real-value ESPRIT algorithm is proposed by [62] and [63], which is on the foundation of
the Unitary ESPRIT, by constructing a transformation matrix, transforms the complex data
of original array into real-value data. Thus lowered the computational burden. Moreover
this algorithm is also applicable to centro-symmetric sensor arrays.
This chapter bases on the foundation of the algorithm that above references proposes and
reference [64], analyzes the rotational invariance principle of RVS-ESPRIT algorithm, and
the relationship of RVS-ESPRIT and complex space ESPRIT(CS-ESPRIT), definitely give:
1. The rotational invariance relationship of the real-value space array steering,
2. The rotational invariance relationship of the real-value space signal subspace,
3. The rotational invariance relationship between the array steering and the signal
subspace of the real-value space,
4. The rotational invariance relationship between the real-value space signal subspace and
the complex value space signal space,
5. The rotational invariance relationship between the real-value space array steering and
the complex value space array steering.
And give the implementing algorithm of REV-ESPRIT. At last compares its performance
with other algorithm by simulation.


128
This chapter is organized as follows
[65]
. It starts with a review of the signal model and the
rotational invariance subspace principle. Next the RVS-ESPRIT algorithm is analyzed,
among which includes the transformation from the complex space to real-value space, the
rotational invariance principle of real-value space, and its implementing algorithm. Finally,
the computer simulations with the comparison the performance of RVS-ESPRIT and the
well-known LS- ESPRIT algorithm are given.
4.1 Signal model
Assume that there are two completely same subarray, their space is already known, and
every subarray consists of m elements. Consider N ( ) N m < narrowband plane waves
from far-field of the array, these plane waves are assumed to be impinging on the array
from directions
1 2
, , ,
N
" , among them,
i
, 1, 2, , i N = " is angle between the
array normal and the direction of the ith signal of the N narrowband planes waves imping.
Because the structure of two arrays is completely same, therefore, for a signal, the difference
of the two subarray outputs is only one phase difference
i
, 1, 2, , i N = " . Suppose
the first subarray receives the data for
1
X , the second receives the data for
2
X , then:
( ) ( )
1 1 1 1 N
= + = +

X a a S N A S N " (4.1)
( ) ( )
1
2 1 2 2
N
j j
N
e e

= + = +

X a a S N A S N " (4.2)
Where, the direction matrix of subarray 1 is ( ) ( )
1 1 N
= =

A A a a " , the direction
matrix of subarray 2 is
2
= A A, S is the space signal vector,
1
N and
2
N are the noise
vectors of the subarray 1 and 2, respectively,and are assumed to be white Gaussian, and
among the formula:

1 N
j j
diag e e

=

" (4.3)
4.2 The rotational invariance subspace principle
From the above mathematics model, we can know that the signal direction information is
included in A and , because is a diagonal matrix, so that we can obtain the DOA of
signal through solving , that is:

2 sin
k
k

= (4.4)
where is the center wave-length of Arriving the wave. So if we obtain the rotational
invariance relationship of the two subarray, we can get the signal DOA information.
First uniting the two subarray models, namely:

1 1
2 2

= = + = +

X A N
X S A S N
X A N
(4.5)
Under the ideal condition, the covariance matrix is estimated as fellows:


129

{ }
H
H
S N
E = = + R X X A R A R
(4.6)
where
{ }
H
S
E = R S S ,
{ }
H
N
E = R N N .
Let the eigendecompositions of the covariance matrix, there is:

2
1
m
H H H
i i i S S S N N N
i
e e
=
= = +
R U U U U
(4.7)
Very obviously, the eigenvalue that gets from the top have the relationship as follows:
1 1 2 N N m

+
> = = " " . where
S
U is signal subspace that spanned by eigenvectors
which are corresponding to large eigenvalues,
N
U is noise subspace that spanned by
eigenvectors which are corresponding to small eigenvalues.
We know that the signal subspace is spanned by large eigenvector is equal to that is
spanned by array direction matrix in the above eigendecomposition, that is:

{ } ( )
{ } S
span span = U A
(4.8)
At this time, existing a nonsingular matrix T , which can make:

( )
S
= U A T
(4.9)
Obviously above-mentioned structure is coming into existence to the two subarrays, so
have:

1
2
S
S
S

= =

U A T
U
U A T
(4.10)
Very obvious, the subspace spanned by array direction matrix A is equal to
1 S
U and
2 S
U
which are spanned by the large eigenvectors of subarray 1 and 2 respectively.

{ } ( ) { } { }
1 2 S S
span span span = = U A U
(4.11)
Moreover, from the relationship of the two subarrays with regard to signal direction matrix,
we can know:

2 1
= A A (4.12)
Again from (4.10),we can know:

1
1 1
1
2
2 1
S S
S
S S
= =

=
= =
U A T A U T
U A T
U A T U T T
(4.13)

1
2 1 1 S S S
= = U U T T U (4.14)
where
1
= T T . (4.12) reflects the rotational invariance characteristic of the signal
direction matrix of the two subarrays, but (4.14) reflects the rotational invariance
characteristic of the received signal data subspace of the two subarrays.


130
If the signal direction matrix A is full rank, we can obtain form (4.14) as fellows:

1
= T T (4.15)
So that,the diagonal matrix which is consisted of the eigenvalues of
certainly be equal to
, but the every column of T is the eigenvectors of . Therefore, once we get the
rotational invariance matrix , we can obtain the signal DOA from (4.4) directly.
4.3 Real-value space ESPRIT algorithm
4.3.1 The transformation from complex space into realvalue space
We know that the uniform linear array is centro-symmetric, and its signal direction matrix
satisfy the nether formula:

M

= J A A (4.16)
where,
M
J is the M M exchange matrix with ones on its antidiagonal and zeros elsewhere,
and the signal direction matrix makes reference to the first element of the array, the diagonal
matrix
( ) 1 M
= , and the is expressed as (4.3). If the reference point is selected as the
central point of the array, so we have:
( ) ( )
1 2
1 C C C N
= =

A A a a " (4.17)
where

( )
( )
( )
1 1
1
2 2
1
i i
i i
M M
j j T
j M j
C i i
e e e e

= =

a a "
(4.18)
If matrix Q satisfying:

M

= J Q Q (4.19)
we call it as the left real transformation matrix.
For example, Q can be chosen for arrays with an even and odd number of sensors
respectively as the following sparse matrices:

2
1
2
n n
n
n n
j
j

=

I I
Q
J J
(4.20)

2 1
1
2
2
n n
T T
n
n n
j
j
+

=

I 0 I
Q 0 0
J 0 J
(4.21)
Moreover, from the bidirectional averaging algorithm, we can process the array data by
once bidirectional averaging, and insert (4.16) into it, we can obtain:

( )
1
2
FB M M
= + R R J R J
(4.22)


131
Insert
H
S N
= + R A R A R into(4.13), we can obtain:

( )
( )
1
2
1
2
H H
FB S N M S N M
H T
S N M S M M N M

= + + +

= + + +
R A R A R J A R A R J
A R A R J A R A J J R J
(4.23)
because of
( )
( )
H
H
T H H
M M M

= = = J A A J A A A J A , and insert it into
(4.23), get the result:

( ) ( )
( )
'
1 1
2 2
1
2
1
2
H H
FB S S N M N M
H H
S S N
H
L

= + + +
= + +
=
R A R R A R J R J
A R R A R
Z Z
( )
( )
4.24
4.25

where

M L

=

Z X J X J (4.26)
Since:

( )
( )
( )
( ) ( )
1 1
2 2
1
2
1
2
1
2
1 1 1
2
H
H
M L M L
H H T H
M L L M
H T H
M M
H H H
M M
H H H
M M
L L
L
L
L
L L

=

= +
= +

= +

= +

Z Z X J X J X J X J
XX J X J J X J
XX J X X J
XX J XX J
X X J X X J
(4.27)
Because
( )
1
H
L
= R X X is the estimating formula of R. Thus (4.25) is established. When the

row number of data vector X is odd, we can definite:

1
2
T
M L

=

X
X x
X
(4.28)
If we process Z which is defined by (4.26) By means of matrix Q which is defined by (4.20)
or (4.21) as fellows:
( )
2
H
M L
= T X Q Z Q


132

( )
{ } { }
{ } { }
{ } { }
2
1 2 1 2
1 2 1 2
Re Im
2 Re 2 Im
Im Re
H
M L
T T

=

+

=

+

T X Q Z Q
X JX X JX
x x
X JX X JX
(4.29)
If the row dimension of the data vector is even, the transformation matrix is:

( )
{ } { }
{ } { }
2
1 2 1 2
1 2 1 2
Re Im
Im Re
H
M L

=

+

=

+

T X Q Z Q
X JX X JX
X JX X JX
(4.30)
What to need to be noticed here is, the matrix Q which defined by (4.20) and (4.21) satisfies

H
= Q Q I (4.31)
From the transformation relationship of (4.28) and (4.29), we can see that ( ) T X transforms
complex data into real data, so that the computational burden is lowered greatly, and we
can obtain:

( ) ( )
( )
( )
2 2 2 2
1
2
1 1
2 2
1 1
2 2
H
T
H
H H H H H
M L M L M L L M
H H H H
M M M M
H
M FB M
L
L L
L L
=
= =

= =

=
R T X T X
Q Z Q Q Z Q Q Z Q Q Z Q
Q Z Z Q Q Z Z Q
Q R Q
(4.32)
If the eigendecompositions of
FB
R as follows:
[ ]
H
S
FB S N
H
N

=

U
R U U
U
(4.33)
Insert (4.33) into (4.32), we can obtain:
[ ]
H
S H
T M S N M
H
N

=

U
R Q U U Q
U
(4.34)
(4.34) shows that the signal subspace of the transformation matrix
T
R is:

H
S M S
= E Q U (4.35)
Insert (4.24) into (4.32), we can obtain:


133

( )
( )
( ) ( ) ( )
( )
'
'
'
'
1
2
1
2
1
2
1
2
H H H H
T M FB M M S S N M
H H H H
M S S M M N M
H
H H H H
M S S M M N M
H H H
T S S T M N M

= = + +

= + +
= + +
= + +
R Q R Q Q A R R A R Q
Q A R R A Q Q R Q
Q A R R Q A Q R Q
A R R A Q R Q
(4.36)
Therefore, the relationship between the real-value transformed signal direction matrix
T
A
and the original complex signal direction matrix A is given by:

H
T M
= A Q A (4.37)
4.3.2 The real-value space rotational invariance principle
We analyze the signal subspace relationship of the two subarray data in the rotational
invariance subspace algorithm theory, which is given by (4.14)
2 1 S S
= U U . If the array is
uniform linear array, and the overlap element of the two subarrays is maximum, namely,
1 m M = , so the signal subspace rotational invariance of the two subarray data can be
expressed as:

2 1 S S
= K U K U (4.38)
where
S
U is the signal subspace of the received data of the whole uniform linear array, and:

[ ]
( )
1 1
1
0
M
M M

= K I
(4.39)

[ ]
( )
2 1
1
0
M
M M

= K I
(4.40)
In the same way, the rotational invariance of the two subarray signal direction matrix can be
given as follows:

2 1
= K A K A (4.41)
where A is the signal direction matrix of the whole array.
From the definition of (4.39) and (4.40), we can see that
1
K and
2
K satisfies:

1 2 m M
= K J K J (4.42)
Utilize the relationship of the definition (4.19):
M M

= = J Q Q J Q Q again, we can
obtain:

( )
( )
( ) ( )
( ) ( )
( )
2 2 2
1 1 1
1
H
H H H
m M m m m M M M m m m M M M
H
H
H
m m M M m M m M
H
m M
= =
= = =
=
Q K Q Q J J K J J Q J Q J K J J Q
J Q K J Q Q K Q Q K Q
Q K Q
(4.43)


134
therefore, define:

( )
{ } 1 1 2 1 2 2
2Re
H H H H
m M m M m M m M
= + = + = H Q K Q Q K Q Q K K Q Q K Q
(4.44a)
( )
{ } 2 1 1 1 2 2
2Im
H H H H
m M m M m M m M
j j j
= = = H Q K Q Q K Q Q K K Q Q K Q
(4.44b)
so that:

( )
1 1 2
1
2
H
m M
j = Q K Q H H
(4.45a)

( )
2 1 2
1
2
H
m M
j = + Q K Q H H
(4.45b)
From the result given by (4.37):
H
T M M T
= = A Q A A Q A , and insert it into the formula
defined by (4.41):
2 1
= K A K A ,we can obtain the results as follows:

2 1 M T M T
= K Q A K Q A (4.46)
The both side of the upper formula multiplies by the
H
m
Q together, we can obtain:

2 1
H H
m M T m M T
= Q K Q A Q K Q A (4.47)
Using (4.45), and removing the constant factor 1 2 , we can obtain that:
( ) ( )
1 2 1 2 T T
j j + = H H A H H A (4.48)
Via moving item, combination and so on simplifications, we will have:
( ) ( )
1 2 T T
j = + H A I H A I (4.49)
From the definition of (4.3)
1 N
j j
diag e e

=

" again, (4.49) can be simplified as
( ) ( )
1
2 1 1
1
T T T T
j

= + = H A H A I I H A (4.50)
where

( ) ( )
{ }
1
1
1
1
1
1
1
1 1 1
1 1
1 1
1 1 1
1 1
tan tan
2 2
N
N
N
N
T
j j
j j
j j
j j
N
j
diag e e diag
j e e
e e
diag
j e e
diag

= +

=

+ +

=

+ +

=

I I
" "
"
"

( )
( )
4.51
4.52


135
So that, (4.50) reflects the rotational invariance relationship of the real-value space array
steering, but (4.51) reflects the rotational invariance relationship between the real-value
space array steering and the complex value space array steering.
Resembling the derivation of (4.50), from
H
S M S S M S
= = E Q U U Q E , and insert it into the
formula given by (4.38):
2 1 S S
= K U K U , we can obtain:

2 1 M S M S
= K Q E K Q E (4.53)
The both side of the upper formula multiplies by the
H
m
Q together, we can obtain:

2 1
H H
M M S M M S
= Q K Q E Q K Q E (4.54)
Using (4.45), and removing the constant factor 1 2 , we can obtain that:
( ) ( )
1 2 1 2 S S
j j + = H H E H H E (4.55)
Via moving item, combination and so on simplifications, we will have:
( ) ( )
2 1 S S
j + = H E I H E I (4.56)
( ) ( )
1
2 1 1 S S S T
j

= + = H E H E I I H E (4.57)
where
( ) ( )
1
T
j

= + I I (4.58)
So that, (4.57) reflects the rotational invariance relationship of the real-value space signal
subspace, but (4.58) reflects the rotational invariance relationship between the real-value
space signal subspace and the complex value space signal space.
Utilizing the character that the space spanned by array direction matrix is equal to which is
spanned by the signal subspace, so a nonsingular matrix
T
T exists, and satisfying
T S T
= A E T , thus using (4.50):
2 1 T T T
= H A H A , we can obtain that:

1
2 1 2 1 S T S T T S S T T T
= = H E T H E T H E H E T T (4.59)
Comparing with (4.57), we can obtain that:

1
T T T T
= T T (4.60)
This formula reflects the rotational invariance relationship between the array steering and
the signal subspace of the real-value space.
4.3.3 The real-value space ESPRIT algorithm
The observational data of M elements are given as:
( ) ( )
1
, ,
M
x t x t " , 1, t L = "


136
Step 1. Construct the M L observational data matrix ( ) ( ) 1 , , L =

X x x " , where
( ) ( ) ( )
1
, ,
T
M
t x t x t =

x " is the observational data vector which is consists of
M elements observational signals.
Step 2. Get the estimating formula of R by
( )
1
H
L
= R X X , and transform the received

array data into real-value space T
R via (4.32).
Step 3. Compute the eigendecompositions of the real-value space T
R , and get the signal

subspace S
E , and the source number N
.
Step 4. Solve the rotational invariance of (4.57) by least square method (or total least square
method), and gain T
.
Step 5. Compute the eigendecompositions of T
, where
1
T T T T

= T T , get
{ } 1
, , T
N
diag
= " .
Step 6. If T
is the real diagonal matrix, according as (4.3) and (4.52), compute the DOA of
imping signal as fellows:

( ) 2 arctan
1, ,
arcsin
2
k k
k k
k N

=

=

" (4.61)
If
k
1, , k N

=

" is complex, compute the DOA by (4.61) with the real part of
k
.
4.4 Simulations
In order to validating the correctness and the effective of the proposed algorithm, we
present some simulation results to illustrate the performance of RVS-ESPRIT. We consider a
ULA with M=8 element and the interelement space is equal to a half of wavelength. There
are three signals impinge on the array from
1
80 = ,
2
20 = ,
3
40 = . The detailed
simulation results are shown as Fig. 26. ~ Fig. 29.
Fig. 26. and Fig. 27. depicts DOA departure versus snapshot number results of RVS-ESPRIT
and TLS-ESPRIT respectively, where the SNR=5dB. In figure 26. and 27., the x-axis denotes
the snapshot number, and y-axis denotes the departure of signal DOA.
Fig. 28. and Fig. 29. depicts DOA departure versus SNR results of RVS-ESPRIT and TLS-
ESPRIT respectively, where the snapshot number =1000. In figure 28. and 29., the x-axis
denotes the SNR, and y-axis denotes the departure of signal DOA.
From the detecting results and comparison between RVS-ESPRIT and TLS-ESPRIT, we can
conclude that RVS-ESPRIT can detect DOA of signal quickly and effectively. At the same
time, the results validate the correctness and effective of this algorithm.


137
2.5
2
1.5
1
0.5
0
D
O
A

D
e
p
a
r
t
u
r
e
RS ESPRIT
-80
-20
40
0 200 400 600 800 1000
Snapshot Number

0.5
0.4
0.3
0.2
0.1
0
D
O
A

D
e
p
a
r
t
u
r
e
TLS ESPRIT
-80
-20
40
0 200 400 600 800 1000
Snapshot Number



138

25
20
15
10
5
0
D
O
A

D
e
p
a
r
t
u
r
e
RS ESPRIT
-80
-20
40
-10 5 0 5 10
SNR

Fig. 28. DOA departure versus SNR. Signal DOA=[-80 -20 40], Snapshot number =1000
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
D
O
A

D
e
p
a
r
t
u
r
e
TLS ESPRIT
-80
-20
40
-10 -5 0 5 10
SNR

Fig. 29. DOA departure versus SNR. Signal DOA=[-80 -20 40], Snapshot number =1000


139
4.5 Conclusion
This chapter carrys on the detailed theories analysis of RVS-ESPRIT based on the theory of
CS-ESPRIT, and gives the concrete implementing algorithm. Because the eigendecompositions
of RVS-ESPRIT is in real domain, so the calculation speed is raised consumedly, then the
speed of DOA estimating is improved largely also. Due to the inherent forward-backward
averaging effect, RVS-ESPRIT can separate two completely coherent sources and provides
improved estimates for correlated signals.
5. References
[1] J. Capon. High resolution frequency-wavenumber spectrum analysis. Proc. IEEE, Vol.57,
pp: 1408-1418, Aug.1969.
[2] J. S. Reed, J. D. Mallet, L. E. Brennan. Rapid convergence rate in adaptive arrays. IEEE
Trans. Aerosp. Electron. Syst., Vol. AES-10, No.6, pp: 853-863, Nov.1974.
[3] D. H. Johnson, D. E. Dudgeon. Array signal processing: Concepts and Techniques.
Englewood Cliffs, NJ: Prentice-Hall, 1993.
[4] H. Krim, M. Viberg. Two decades of array signal processing research. IEEE Signal
Process. Mag., Vol.13, No.4, pp: 67-94, Jul.1996.
[5] L. C. Godara. Application of antenna arrays to mobile communications, Part II: Beam-
forming and direction-of-arrival considerations. Proc. IEEE, Vol.85, No.8, pp: 1195-
1245, Aug.1997.
[6] H. L. Van Trees. Detection, estimation, and modulation theory, Part IV, Optimum array
processing. New York: Wiley, 2002.
[7] P. S. Naidu. Sensor array signal processing. Boca Raton, FL: CRC, 2001.
[8] J. R. Guerci. Space-time adaptive processing. Norwood, MA: Artech House, 2003.
[9] Jian Li, Petre Stotica. Robust adaptive beamforming. New York: Wiley, 2006.
[10] H. Cox. Resolving power and sensitivity to mismatch of optimum array processors.
J. Acoust. Soc. Amer., Vol.54, No.3, pp: 771-785, Mar.1973.
[11] D. D. Feldman, L. J. Griffiths. A projection approach to robust adaptive beamforming.
IEEE Trans. Signal Processing, Vol.42, pp: 867-876, Apr.1994.
[12] M. Wax, Y. Anu. Performance analysis of the minimum variance beamformer in the
presence of steering vector errors. IEEE Trans. Signal Processing, Vol.44, pp: 938-
947, Apr.1996.
[13] E. K. Hung, R. M. Turner. A fast beamforming algorithm for large arrays. IEEE Trans.
Aerosp. Electron. Syst., Vol.AES-19, pp: 598-607, Jul.1983.
[14] B. D. Carlson. Covariance matrix estimation errors and diagonal loading in adaptive
arrays. IEEE Trans. Aerosp. Electron. Syst., Vol.24, pp: 397-401, Jul.1988.
[15] M. Wax, Y. Anu. Performance analysis of the minimum variance beamformer. IEEE
Trans. Signal Processing, Vol.44, pp: 928-937, Apr.1996.
[16] H. Cox, R. M. Zeskind, M. H. Owen. Robust adaptive beamforming. IEEE Trans.
Acoust., Speech, Signal Processing, Vol.ASSP-35, pp:1365-1376, Oct.1987.
[17] A. B. Gershman. Robust adaptive beamforming in sensor arrays. Int. J. Electron.
Commun., Vol.53, pp: 305-314, Dec.1999.
[18] A. B. Gershman. Robust adaptive beamforming: an overview of recent trends and
advances in the field. International Conference on Antenna Theory and Technique,
9-12 September, 2003, Sewstopol, Ukraine, pp: 30-35.


140
[19] Sergiy A. Vorobyov, A. B. Gershman, Zhi-Quan Luo. Robust adaptive beamforming
using worst-case performance optimization: a solution to the signal mismatch
problem. IEEE Trans. Signal Processing. Vol.51, No.2, pp: 313-324, Feb.2003.
[20] Jian Li, Petre Stotica, Zhisong Wang.On robust capon beamformer and diagonal
loading. IEEE Trans. Signal Processing. Vol.51, No.7, pp: 1702-1715, Jul.2003.
[21] S. Shahram, A. B. Gershman, Zhiquan Luo, et al. Robust adaptive beamforming for
general-rank signal models. IEEE Trans. Signal Processing, Vol.51, No.9, pp: 2257-
2269, Sep.2003.
[22] Jian Li, Petre Stotica, Zhisong Wang, Doubly constrained robust capon beamformer.
IEEE Trans. Signal Processing. Vol.52, No.9, pp: 2407-2423, Sep.2004.
[23] G.L. Robert, P.B. Stephen. Robust minimum variance beamforming. IEEE Trans. Signal
Processing, Vol.53, No.5, pp: 1684-1696, May.2005.
[24] Ayman Elnashar, Said M. Elnoubi, Hamdi A. El-Mikati. Further study on robust
adaptive beamforming with optimum diagonal loading. IEEE Trans. Antennas
Propagation, Vol.AP-54, No.12, pp: 3647-3658, Dec.2006.
[25] C.Y.Chen, P.P.Vaidyanathan. Quadratically constrained beamforming robust against
direction-of- arrival mismatch. IEEE Trans. Signal Processing, Vol.55, No.8, pp:
4139-4150, Aug.2007.
[26] R. A. Monzingo, T. W. Miller. Introduction to adaptive arrays. New York: Wiley, 1980.
[27] Z. Tian, K. L. Bell, H. L. Van Trees. A recursive least squares implementation for LCMP
beamforming under quadratic constraint. IEEE Trans. Signal Processing, Vol.49,
No.6, pp: 1365-1376, Jun.2001.
[28] L. C. Godara. Error analysis of the optimal antenna array processors. IEEE Trans.
Aerosp. Electron. Syst., Vol. AES-22, pp: 395-409, Jul.1986.
[29] K. L. Bell, Y. Ephraim, H. L. Van Trees, A Bayesian approach to robust adaptive
beamforming. IEEE Trans. Signal Processing, Vol.48, pp: 386-398, Feb.2000.
[30] L. Chang, C. C. Yeh. Performance of DMI and eigenspace-based beamformers. IEEE
Trans. Antennas Propagation, Vol.40, pp: 1336-1347, Nov.1992.
[31] Cheng-Chou Lee, Ju-Hong Lee. Eigenspace-based adaptive array beamforming with
robust capabilities. IEEE Trans. Antennas Propagation, Vol.45, pp: 1711-1716,
Dec.1997.
[32] J. Riba, J. Goldberg, G. Vazquez. Robust beamforming for interference rejection in
mobile communications. IEEE Trans. Signal Processing, Vol.45, pp: 271-275,
Jan.1997.
[33] J. R. Guerci. Theory and application of covariance matrix tapers for robust adaptive
beamforming. IEEE Trans. Signal Processing, Vol.47, pp: 997-985, Apr.1999.
[34] J. R. Guerci, J.S.Bergin. Principal components,covariance matrix tapers,and the subspace
leakage problem. IEEE Trans. Aerosp. Electron. Syst, Vol.38, pp: 152-162, Jan.2002.
[35] F. Vincent, O. Besson. Steering vector errors and diagonal loading, IEE Proceedings.-
Radar Sonar Navig., Vol.151, No.6, pp: 337-343, Dec.2004.
[36] O. Besson, F. Vincent. Performance analysis of beamformers using generalized loading
of the covariance matrix in the presence of random steering vector errors. IEEE
Trans. Signal Processing, Vol.53, pp: 452-459, Feb.2005.
[37] Almir Mutapcic, Seung-Jean Kim, Stephen Boyd. Beamforming with uncertain weights.
IEEE Signal Processing Letter, Vol.14, No.5, pp: 348-351, May.2007.


141
[38] Liu Congfeng, Liao Guisheng, Robust capon beamformer under norm constraint, Signal
Processing, May.2010, Vol.90, Issue.5, pp: 1573-1581.
[39] Dolph C L. A current distribution for broadside arrays which optimizes the relationship
between beam width and sidelobe level. Proc IRE, June.1946, Vol.34, pp:335348.
[40] Zhou P, Ingram M. Pattern synthesis for arbitrary arrays using an adaptive array
method. IEEE Transactions on Antennas And Propagation. May.1999, Vol.47, No.5,
pp:862-869.
[41] Wang F, Balakrishnan V, Zhou P Y, et al. Optimal array pattern synthesis using
semidefinite programming. IEEE Transactions on Signal Processing, May.2003,
Vol.51, No.5, pp:1172-1183.
[42] Guo Q, Liao G., Wu Y, et al. Pattern synthesis method for arbitrary arrays based on
LCMV criterion. Electronics Letters. May.2003, Vol.39, No.23, pp:1628-1630.
[43] Xie Yao, Jian Li, Zheng Xiayu, et al. Optimal array pattern synthesis via matrix
weighting, ICASSP, IEEE International Conference on Acoustics, Speech and Signal
Processing Proceedings, Apr 15-20 2007, Honolulu, HI, United States, IEEE Inc.,
Piscataway, NJ 08855-1331, United States, Vol.2, pp: 885-888.
[44] Kurup D G., Himdi M, Rydberg A. Synthesis of uniform amplitude unequally spaced
antenna arrays using the differential evolution algorithm, IEEE Transactions on
Antennas And Propagation. Sep.2003, Vol.51, No.9, pp:2210-2217.
[45] Boeringer D W, Werner D H. Particle swarm optimization versus genetic algorithms for
phased array synthesis, IEEE Transactions on Antennas And Propagation.
Mar.2004, Vol.52, No.3, pp:771-779.
[46] Liu Xiaojun, Liu Congfeng, Liao Guisheng, Improved pattern synthesis method with
linearly constraint minimum variance criterion, 2010 IEEE Internation Conference
on Wireless Communications, Networking and Information Security (WCNIS2010),
2010.6,25-27, Vol.2, Beijing China
[47] Barabell A.J, Improving the resolution performance of eigenstructure-based direction-
finding algorithms.Proc,ICASSP 83,1983,pp.336-339.
[48] Kumaresan R. and Tufts D.W., Estimating the angles of arrival of multiple plane
wave,IEEE Trans,1983,AES-19,pp.134-138.
[49] Bao B and Hari.K, Performance analysis of root-MUSIC, IEEE Trans, 1989, ASSP-
37,pp.1939-1949.
[50] Li F and Vaccaro.R.J,Analytical performance prediction of subspace-based algorithms
for DOA estimation: SVD and Signal Processing II Algorithms,Analysis and
Applications (Elsevier,New York,1991)
[51] Xu X.L and Buckley K.M, Reduced dimension beamspace broad-band localization,
preprocessor design and evaluation.Proc.IEEE Forth Workshop on Spectrum
Estimation and Modeling, 1988,pp22-26.
[52] Ta sung Lee, Fast implementation of root-form eigen-based methods for detecting
closely spaced sources IEE Proceedings-F Vol.139.No.4,August 1992.
[53] Marius P, Alex B.G and Martin H, Unitary root-MUSIC with a real-valued
eigendecomposition: a theoretical and experimental performacce study, IEEE
Transactions on SP,Vol.48,No.5,May 2000,pp:1306-1314.
[54] Liu Congfeng, Liao Guisheng, Fast algorithm for root-MUSIC with real-valued
eigendecomposition, Radar-2006: 2006 CIE International Conference on Radar
2006.10 ShangHai China.


142
[55] Q. S Ren and A. J. Willis, Extending MUSIC to single snapshot and on line direction
finding applications, International Radar Conference, IEE, Stevenage, United
Kingdom, Oct.1997: 783-787.
[56] Liu Xiaojun, Liu Congfeng, Liao Guisheng, Polynomial coefficient finding for root-
MUSIC, Journal of Electronic (China), 2009.8, Vol.26(4).
[57] R. Roy and T. Kailath, ESPRIT-Estimation of signal parameters via rotational invariance
techniques, in Signal Processing Part 11: Control Theory and Applications (L.
Auslander, F. A. Griinbaum, J. W. Helton, T. Kailath, P. Khargonekar, and S. Mitter,
Eds.). Berlin, Vienna, New York Springer-Verlag, 1990, pp. 36941 1.
[58] R. H. Roy, ESPRIT-estimation of signal parameters via rRotational invariance techniques,
Ph.D. thesis, Stanford Univ., Stanford, CA, Aug. 1987.
[59] M. Haardt and M. E. Ali-Hackl, Unitary ESPRIT: How to exploit additional
information inherent in the rotational invariance structure, in Proc. IEEE Int. ConJ
Acoust., Speech, Signal Processing, Adelaide, Australia, Apr. 1994, pp. 229-232, vol.
IV.
[60] S. U. Pillai and B. H. Kwon, Forwardhackward spatial smoothing techniques for
coherent signal identification, IEEE Trans. Acoust., Speech, Signal Processing, vol.
37, pp. 8-15, Jan. 1989.
[61] B. D. Rao and K. V. S. Hari, Weighted subspace methods and spatial smoothing:
Analysis and comparison, IEEE Trans. Signal Processing, vol. 41, pp. 788-803, Feb.
1993.
[62] Martin Haardt, Josef A. Nossek Unitary ESPRIT: how to obtain increased estimation
accuracy with a reduced computational burden IEEE Trans. Signal Processing, Vol.
43, No. 5, pp.1232-1242. MAY 1995
[63] B. Ottersten, M. Viberg, and T. Kailath, Performance analysis of the total least squares
ESPRIT algorithm, IEEE Trans. Signal P mcessing, vol. 39, pp. 1122-1135, May
1991.
[64] Wang Yongliang, Chen Hui, Peng Yingning, Wan qun Spatial spectrum estimation
techniques and algorithm Tsinghua university press Beijing China 2004.
[65] Liu Congfeng, Liao Guisheng, Real-value space ESPRIT algorithm and Its implement,
WiCOM 2006: 2006 International Conference on Wireless Communications,
Networking, and Mobile Computing 2006.09 WuHan China.
1. Introduction
The solar corona has a very complex and highly dynamic structure. It consists of a large
number of constantly evolving, loops and laments, which interact with each other and
are closely associated with the local magnetic eld. The non-stationary character of solar
plasma-magnetic structures manifests itself in various forms of the coronal magnetic loops
dynamics as rising motions, oscillations, meandering, twisting (Aschwanden et al., 1999;
Schrijver et al., 1999), as well as in formation, sudden activation and eruption of laments
and prominences. Energetic phenomena, related to these types of magnetic activity, range
from tiny transient brightenings (micro-ares) and jets to large, active-region-sized ares
and coronal mass ejections (CMEs). They are naturally accompanied by different kinds
of electromagnetic (EM) emission, covering a wide frequency band from radio waves to
gamma-rays. Radiation, produced within a given plasma environment, carries an information
on physical and dynamic conditions in a radiating source. This causes an exceptional
importance of the EMradiation, as a diagnostic tool, for understanding the nature and physics
of various solar dynamic phenomena. As a relatively new, in that context, direction of study
in the traditional branch of the solar microwave radio astronomy appears the analysis of the
slow, long-periodic (e.g., > 1 s) uctuations of the radiation intensity (Khodachenko et al.,
2005; Zaitsev et al., 2003).
Microwave radiation from the magnetic loops in solar active regions (e.g., during solar
ares) is usually interpreted as a gyro-synchrotron radiation, produced by fast electrons on
harmonics of the gyro-frequency
B
in the magnetic eld B of the loop. In the case of a
power-lawdistribution of electrons in energy as f (E) E
, the intensity of gyro-synchrotron

radiation I
from an optically thin loop (Dulk, 1985; Dulk & Marsh, 1982) is
I
B
0.22+0.9
(sin )
0.43+0.65
, (1)
where is the angle between magnetic eld and the direction of electromagnetic wave
propagation. For the observed typical values of the electron energy spectrum index 2 7
this implies the proportionality of intensity to a moderately high power of the background

Analysis of Long-Periodic Fluctuations of Solar
Microwave Radiation, as a Way for Diagnostics
of Coronal Magnetic Loops Dynamics
Maxim L. Khodachenko
1
, Albert G. Kislyakov
2
and Eugeny I. Shkelev
2
1
Space Research Institute, Austrian Academy of Sciences, Graz,
2
Lobachevsky State University, Nizhny Novgorod,
1
Austria
2
Russia
5
magnetic eld and essential anisotropy of the radiation: I
B
1.586.08
( sin )
0.874.12
.
Equation (1) is obtained within an assumption of an optically thin source, when the radiation
intensity is proportional to the emissivity
(Dulk, 1985). According to the estimations in

Urpo et al. (1994), a coronal loop with diameter of about 4 10
8
cm is optically thin in the
considered frequency range for the gyro-synchrotron absorption if the density of fast electrons
is < 2 10
9
cm
3
. The typical density of > 10 keV electrons in the microwave burst events is
usually 10
6
10
7
cm
3
and therefore, it stays well within the above indicated limit.
It follows from the equation (1) that variations of the loop magnetic eld, associated with
disturbances of the electric current in a radiating source, should modulate the intensity of
the microwave radiation of the loop (Khodachenko et al., 2005; Zaitsev et al., 2003). Another
origin for the modulation of intensity of the observed microwave radiation can be due to the
quasi-periodic motion (oscillation) of a coronal magnetic loop, containing the radiation source.
This mechanismis connected with the anisotropy of the gyro-synchrotron emission, as well as
with the variation of the magnetic eld value during the oscillatory motion of the loop, which
both, according to the equation (1), can result in a quasi-periodic modulation of the received
signal (Khodachenko et al., 2006; 2011). Therefore, the analysis of slow modulations of solar
microwave radiation may be used for the diagnostics of oscillating electric currents in the
coronal loops, as well as for the investigation of large-scale motion of the loops (including loop
oscillations) in solar active regions. By this, it is natural to expect that structural complexity of
solar active regions will manifest itself in peculiarities of the emitted radiation.
The dynamic spectra of the long-periodic oscillations, modulating the intensity of microwave
radiation from solar active regions, have been found to contain quite often several spectral
tracks, demonstrating a specic temporal behaviour (Zaitsev et al., 1998; 2001a;b; 2003).
Khodachenko et al. (2005) considered these multi-track features as an indication that the
detected microwave radiation is produced within a system of a few closely located, magnetic
loops, having slightly different parameters and involved in a kind of common global
dynamic process. In several cases such slow modulations of solar microwave radiation
(with multi-track spectra) were interpreted as the signatures of oscillating electric currents,
running within the circuits of moving relative each other inductively connected coronal
magnetic loops (Khodachenko et al., 2006). The dynamics of these electric currents has been
described by means of the equivalent electric circuit (LCR-circuit) models of the coronal loops
(Khodachenko et al., 2003; Zaitsev et al., 1998) characterized by time-dependent inductance
L, capacitance C, resistance R, as well as mutual inductance coefcients M
j
(Khodachenko et
al., 2003; 2009). The L, C, R and M
j
parameters of electric circuit of a current-carrying loop
depend on shape, scale, position of the loop with respect to other loops, as well as on the
plasma parameters and value of the total longitudinal current in the magnetic tube. In that
respect it is worth to mention that the LCR-circuit model ignores the fact that changes of the
magnetic eld and related electric current propagate in plasma at the Alfvn speed. It ignores
any short-time variations of plasma parameters, which appear to be averaged in course of
derivation of the LCR model equation (Khodachenko et al., 2009; Zaitsev et al., 2001a). The
LCR approach assumes instant changes of the electric current over the whole electric circuit
according to the varying potential and ignores all the "propagation effects" related to the
system MHD modes. The LCR equations correctly describe temporal evolution of electric
currents in a system of solar magnetic current-carrying loops only at a time scale longer
than the Alfvn wave propagation time. More generally, the equivalent electric circuit model
of a coronal loop tends to emphasize the global electric circuit, obscuring the effects of the
Analysis of Long-Periodic Fluctuations of Solar Microwave Radiation, as a Way for Diagnostics of Coronal Magnetic Loops Dynamics 3
ambient plasma and details of the magnetic structure. The LCR model approach was applied
in particular for interpretation of the solar microwave burst long-periodic modulations with
drifting modulation frequencies (t) in the interval 0.03 1 Hz. Based on the analysis of the
frequency drift of the modulations Zaitsev et al. (1998; 2001a; 2003) and Khodachenko et al.
(2005) estimated also the values of the electric current in aring loops (10
8
10
11
A), which
appeared to be close to the values obtained by other methods (Hardy et al., 1998; Leka et al.,
1996; Moreton & Severny, 1968; Spangler, 2007; Tan et al., 2006).
However, not all long-periodic modulations of solar microwave radiation demonstrate
frequency drift and occupy a similar frequency range. Khodachenko et al. (2009) pointed
out that the frequency of LCR-oscillations of the electric current, which depends on specic
parameters of a coronal loop, usually stays within the interval
LCR
(0.03 1) Hz.
Therefore, the modulations of the solar microwave radiation intensity with 0.03Hz< < 1Hz
are very likely to be due to the electric currents, oscillating in LCR-circuits of coronal loops.
At the same time, modulations caused by the oscillatory motions of loops that contain the
radiation sources, because of their direct connection with the large-scale dynamics of loops,
should have typical frequencies < 0.01 Hz and exhibit no drift. Thus, it has been proposed in
(Khodachenko et al., 2009) that one should distinguish, when speaking about different kinds
of long-periodic modulations of the solar microwave radiation, between the low-frequency (LF)
( 0.03 1 Hz) and very-low-frequency (VLF) (< 0.01 Hz) modulations, assuming the rst to be
connected with the LCR-oscillations of electric currents in the coronal loops and the second to
be caused by large-scale motions of the radiation sources conned within the oscillating loops.
LF modulations (e.g., 0.03 1 Hz) have been studied in details and interpreted in terms of the
equivalent electric circuit models of coronal loops in Khodachenko et al. (2005; 2006); Zaitsev
et al. (1998; 2001a;b; 2003). They are not considered in this chapter, whereas we addresses here
the VLF modulations (< 0.01 Hz) of solar microwave radiation and their possible relation to
the large-scale dynamics of coronal loops in solar active regions. Some preliminary results on
that subject were reported recently in a short publication Khodachenko et al. (2011), and the
present paper addresses this topic in more details.
Transverse oscillations of the coronal magnetic loops considered here are triggered by ares
and lament eruptions, i.e. by phenomena in which the signicant Lorentz forces are likely
acting in association with the magnetic eld adjustments. By this, all considered oscillating
loops have at least one footpoint in the immediate vicinity of a separatrix surface or of a
are ribbon. Using the potential-eld extrapolations Schrijver & Brown (2000) demonstrated
that the eld lines close to a separatrix surface exhibit strongly amplied displacements in
response to small displacements in the photospheric "roots". This means that the magnetic
eld lines in the proximity of separatrix are much more sensitive to changes in the eld sources
than are the eld lines that lie well within domains of connectivity. Speaking about a nature
of the observed transversal oscillations of coronal loops Schrijver et al. (2002) address two
models: (a) transverse waves in coronal loops that act as wave guides and (b) mentioned
above, strong sensitivity of the shape of magnetic eld lines near separatrix to changes in the
bottom eld sources. By this, the authors outline several observational features that favor the
model (b). Based on the extensive study of properties of transverse loop oscillations triggered
by ares, Aschwanden et al. (2002) also concluded that most of the loops do not t the simple
model of a kink eigen-mode oscillation. Therefore, the present paper is not dedicated to
the study of MHD oscillations in solar coronal loops. Our goal consists in demonstration
of the fact that quasi-periodic transverse motions of a coronal magnetic loop, which contains
145
Microwave Radiation, as a Way for Diagnostics of Coronal Magnetic Loops Dynamics
a source of microwave emission, may be connected with a specic modulation of radiation
intensity received by a remote observer.
Speaking about other possible mechanisms (besides of the microwave radiation source
large-scale oscillatory motion), which may cause a quasi-periodic modulation of the
non-thermal electron gyro-synchrotron radiation, it is necessary to mention that a
quasi-periodically varying ow of the non-thermal electrons may also result in oscillations
of intensity of microwave radiation. Generation of energetic electrons usually is believed to
be associated with the processes of magnetic reconnection during solar ares (Miller et al.,
1997). There are also theories which suggest acceleration of particles by the inductive and
charge separation electric elds, build in course of the continuous motion of solar large-scale
coronal magnetic structures (Khodachenko et al., 2003; Zaitsev & Stepanov, 1992). Besides
of that, particle acceleration in a collapsing magnetic trap (Karlick & Kosugi, 2004), in the
MHD turbulence (LaRosa & Moore, 1993; Miller et al., 1996), and in shocks (Cargill et al.,
1988; Holman & Pesses, 1983) are addressed as secondary possible mechanisms for energetic
particle production. An extended review of particle acceleration processes in solar ares was
recently published by Aschwanden (2002). In most of these cases the typical periods are
shorter than those of the VLF transverse oscillations of coronal loops. On the other hand,
there are also models in which VLF large-scale oscillations of coronal loops control the process
of generation of energetic particles after the impulsive phase of a are (Nakariakov et al.,
2006). This case, however, deserves a special study, which appears beyond the scope of the
present paper. Our analysis here is based on the traditional scenario, according to which the
non-thermal particles, produced during a are in particle acceleration regions (e.g., sites of
magnetic reconnection or the area of separatrix currents), are injected into oscillating loops.
2. Peculiarities of VLF modulations of microwave radiation related to large-scale
transverse motions of the radiating sources in the oscillating coronal loops
Even taking into account the typical ranges of the modulation frequencies of microwave
radiation (mentioned above), it is usually difcult to identify the modulation mechanism
acting in each particular case. Indeed, having in mind only Equation (1) one cannot say for
sure if the observed modulation is due to a) the electric current oscillations in the radiation
source or b) the large-scale oscillatory motion of the loop. However, regarding the last
modulation mechanism, an attention should be paid to the fact that large-scale transverse
oscillatory motion of a coronal loop is accompanied by the periodic stress of magnetic eld
created in the loop, especially near its footpoints, during each inclination, i.e. two times
per oscillation cycle. This means that the magnetic eld strength uctuates during the
oscillatory motion of the loop with a half-period P
osc
/2 of the loop oscillation. Therefore,
according to Equation (1), for a transverse oscillating loop, a properly located observer, in
addition to the modulation caused by the emission diagram motion at the main oscillation
frequency
0
= 1/P
osc
, may see in some cases the modulation at the double frequency of the
loop oscillation 2
0
, as well as weak higher order harmonics caused by the non-linearity of
Equation (1). However, as it will be shown below the relative amplitude of the higher-order
harmonics, i.e., with numbers > 2 is rather low, and in the most cases only the rst two
harmonic frequencies can be detected. The domination of the main and double-frequency
harmonics in the spectrum is caused by implicit presence of these frequencies in the signal,
according to the above described character of the radiation modulating factors. Therefore,
the presence of the "modulation pairs" in the low-frequency spectra, i.e., the lines which can
be associated with the main and double frequency of the loop oscillation (
0
and 2
0
) may
indicate about a transverse oscillatory dynamics of the loop.
Formation of the "modulation pairs" and their higher-order harmonic companions in
multi-line dynamic spectrum of the VLF modulation of microwave radiation emitted from
a transverse oscillating coronal loop may be illustrated with a simple model. Lets suppose
that the loop undergoes oscillations in the direction transverse to the loop plane as shown in
Figure 1. The loop inclination relative to the vertical direction varies as (t) =
0
sin(2
0
t),
where
0
and
0
are the angular amplitude and frequency of the loop oscillations, respectively.
Assuming that this loop, when oriented vertically, is seen by a remote observer at the angle
0
,
we get that in course of the loop oscillation the viewing angle changes as (t) =
0
(t).
Irrespectively of the nature of a coronal loop oscillation, the important feature of the
large-scale transverse motion of the loop, consists in an oscillating magnetic stress, created in
the loop during its quasi-periodic inclinations. Assuming the local transverse disturbance of
the magnetic eld relative its initial vertical direction to be B, we nd that the total disturbed
magnetic eld is B/cos (t). For sufciently small (t) the following approximation can
be used: 1/(cos (t)) 1/(1 (t)
2
)
1/2
1 + (1/2)(t)
2
. This means that the disturbed
magnetic eld in the loop varies in time as B(t) B(1 + (1/2)(t)
2
). Therefore, for the
assumed above sinusoidal character of (t), we nally obtain that the local magnetic eld in a
transverse oscillating magnetic loop may be approximated as B(t) (1 +0.5
2
0
sin
2
(2
0
t)).
Substitution of the expressions for (t) and B(t) into (1) enables to construct a modeling
signal for the varying intensity of microwave radiation emitted from a transverse oscillating
magnetic loop. The examples of dynamic spectra of this signal obtained with
0
= /6 and
= 5 for different viewing angles
0
= /2; /3; /4; /6 are shown in Figure 2.
Dynamical spectra of the modeling signal in Figure 2 demonstrate several important features,
typical for the radiation emitted from a microwave source located in a transverse large-scale
oscillating magnetic loop, which may be observed in the solar microwave emission. In
particular, for the most of the viewing angles (except of
0
= /2) the dynamic spectra
contain well pronounced "modulation pairs", e.g. the lines at the main
0
and double 2
0
frequency of the oscillation. Besides of that, sometimes also a weak third harmonic at 3
0
may be observed, which appears due to essentially non-sinusoidal (non-harmonic) character
of the signal resulted from the joint action of two modulating factors: quasi-periodic magnetic
stress and emission diagram motion. In a special case of
0
= /2, the absence of the main
frequency component is caused by a "symmetrizing" (in this case) of the varying angular part
of the emission intensity. This results in a situation when the diagram motion and magnetic
stress factors work synchronously.
As it can be seen in Figure 2, only the rst two harmonics have high enough amplitudes. In
particular, the spectral amplitude of third harmonic in the cases with
0
= /3; /4; /6,
never exceeds 25% of the main frequency component, whereas the second harmonic
constitutes usually about 65% of the last. Therefore, the detection of harmonics with numbers
higher than 2 in a natural signal, will be in the most cases difcult due to the noise
contamination. The presence of modulation pairs in the VLF spectra of solar microwave
radiation may be considered as an imprint of a transverse kink-type motion of a loop
containing the radiation source. This feature may be used for the indirect identication of
candidates for transverse oscillating coronal loops by nding specic modulation lines in the
VLF dynamic spectra of microwave radiation. However, the exact detection of transverse
147
Fig. 1. Schematic view of an oscillating loop which contains a microwave radiation source.
0
is the direction to a remote observer and
0
is the angular amplitude of the loop
oscillations.
motion of the radiating loops by the dynamical spectra of microwave emission (with the
exclusion of other mechanisms which may also generate higher spectral harmonics) requires
quantitative study of the measured radio signal and superimposing these results with the
precise calculation of the radiation fromthe loop, taking into account the loop position relative
observer.
Several real observational examples are considered below, for which VLF modulations of
microwave radiation could be associated with the observed in EUV post-are oscillating
coronal loops.
Fig. 2. Dynamical spectra of the modeling signal with
0
= /6 and = 5 demonstrating the
peculiarities of the microwave radiation VLF modulations produced due to large-scale
transverse oscillations of a coronal loop containing the radiating source. Different directions
to an observer
0
(viewing angles) are considered: (a) /2; (b) /3; (c) /4; (d) /6.
3. Data preparation and analysis methods
Differently to the idealized innite in time analytical modelling signal considered in section 2,
the natural radio emission received from a solar active region with oscillating coronal loop(s)
is essentially time-dependent. It usually begins with an impulsive phase of a solar are and
has duration of only several periods of decaying oscillations of the loop(s). Besides of that,
a signal from a particular oscillating loop is quite often strongly contaminated by interfering
signals from neighboring loops in the active region of interest, as well as by the radiations
emitted from other solar active regions. This complicates the task of detection and diagnostics
of coronal magnetic loop oscillations in microwaves and requires special data preparation
procedures with consequent application of high spectral and time resolution data analysis
techniques.
Since the analyzed data appear in a form of discrete counts of the signal intensity, it is
natural that digital methods are applied for their processing. A basic specics of the digital
methods consists in certain limitation of dynamical range of the resulting spectra which
may lead to the loss of relatively weak and short-time, but important parts of the whole
spectra-temporal picture of the studied phenomenon. To avoid of that, the analyzed data
pass certain pre-processing preparation which (depending on particular case and task) may
include the following procedures: 1) Subtraction of a constant component of a signal, or the
signal average; 2) subtraction of a slow (as compared to the analyzed oscillations) major trend
of the signal; 3) slow polynomial approximation of the analyzed data with the consequent
subtraction of the approximating signal; 4) signal "normalization" (will be described below);
149
and 5) digital ltration of contaminating components. Altogether, the "subtraction" methods
1-3 enhance visibility of weaker uctuations of the radiation, enabling better analysis of their
spectral and temporal characteristics. Note, that the digital ltration (e.g., the method 5) with
an appropriate lter(s) may efciently substitute all the "subtraction" methods. However, the
inertness of a lter results in some smoothing of the radio emission uctuations of interest.
Besides of that, information about the intensity of the radiation uctuations is lost during the
frequency ltration.
The ltration algorithm consists in the application of a Gaussian window to the analyzed
signal spectrum, obtained with the discrete Fourier transform (DFT), with the consequent
performance of the reverse DFT. The Gaussian window for the low frequency (LF), high
frequency (HF), band-pass (BP), and band-lock (BL) lters, respectively, is determined by the
following expressions:
W
LF
(k) = exp
_
k
2
2

_
f
1
f
s
d N
_
2
_
, k = 0, ..., N/2;
W
HF
(k) = 1 exp
_
k
2
2

_
f
1
f
s
d N
_
2
_
, k = 0, ..., N/2;
W
BP
(k) = exp
_
1
2

_
k
N

f
0
f
s
_
2
_
f
f
s
d
_
2
_
, k = 0, ..., N/2;
W
BL
(k) = 1 exp
_
1
2

_
k
N

f
0
f
s
_
2
_
f
f
s
d
_
2
_
, k = 0, ..., N/2.
(2)
Here N is the number of counts in the analyzed signal, f
s
is the frequency of discretization, d
is decimation coefcient (Marple, 1986), f
1
is the cut-off frequency for the LF and HF lters,
f
0
and f , are the central frequency and the band, respectively, for the BP and BL lters.
Lets consider now the "normalization" method. It is based on a treatment of an analytical
signal z(n) (Marple, 1986):
z(n) = s(n) + is
H
(n) = M(n) exp(i(n)), (3)
where s(n) and s
H
(n) are the analyzed digital signal and its Hilbert conjugate, respectively,
and n is the count number. The parameters M(n) and (n) are the module and phase of the
analytical signal, which are dened as the following:
M(n) =
_
s(n)
2
+ s
H
(n)
2
,
(n) = arctan
_
s
H
(n)
s(n)
_
.
(4)
The "normalization" procedure consists in division of the analytical signal z(n) on M(n). By
this, the amplitudes of all spectral components of the analyzed process s(n) become to be
equalized. Note, that due to the orthogonality of functions s(n) and s
H
(n), the value of
function M(n) never becomes zero. The "normalization" method is especially efcient for
the analysis of non-stationary modulation processes. It enables to detect and to follow the
variations of an "instantaneous" frequency of an oscillatory component of radiation.
An important role in the present work belongs also to the method, used for the detection
of quasi-periodic features in the time records of the solar microwave radiation intensity. It
consists in application of an original data analysis algorithm (Shkelev et al., 2002; Zaitsev et
al., 2001b) made as a combination of the "sliding window" Fourier (SWF) transform technique
and the nonlinear Wigner-Ville (WV) method (Cohen, 1989; Ville, 1948; Wigner, 1932). Below
we outline the idea of this data analysis algorithm and its main features.
The classical Fourier transform enables the analysis of a given signal in terms of separate
spectral frequency components. It is applied for the study of relative distribution of energy
between the spectral components in the case of sufciently long (ideally innite) duration
of the analyzed signal. However such energetic spectrum does not provide an information
on a time when each particular spectral component appears. Possible improvement of the
classical Fourier transform method in that respect consists in its application within a certain
interval of time t (so called "window") and in a consequent shift of this "window" along
the time axis. This approach became a standard method for the analysis of non-stationary
signals. Further generalization of SWF transformmethod leads to the wavelet analysis, where
effect of the "window" is produced by means of a certain mother-wavelet function. Wavelet
transform enables judging about energy distribution over the time and frequency of an
analyzed signal. Nowadays this method is also widely used for the analysis of non-stationary
and impulsive signals. Its efciency is however strongly dependent on parameters of the
applied mother-wavelet function, which needs to be specially selected and adjusted to the
type of particular analyzed signal.
Recently one more spectral analysis method has been applied in astrophysics. This method is
based on the Wigner-Ville (WV) transform
P( f , t) =
_

z
_
t +

2
_
z
_
t

2
_
e
i2f
d, (5)
where z(t) = s(t) + is
H
(t) is an analytic signal, made of the analyzed sample of real signal
s(t), and its Hilbert conjugate s
H
(t). Function P( f , t) gives distribution of the signal energy
over frequency f and time, and may be visualized in the form of a dynamical spectrum of the
signal. According to its denition (5), WV transform may be also interpreted as Fourier image
(relative the shifted time) of the local autocorrelation function for the analytical signal z(t).
Since the time t appears explicitly among the arguments of the WV spectrum P( f , t), this
method is most efcient for high-resolution spectra-temporal analysis of non-stationary
signals with varying spectra, such as quasi-harmonic signals with a changing frequency,
or varying impulsive signals. In these cases SWF transform and wavelet methods are less
efcient, because the averaging over the analysis window (or over the wavelet) results in
a decrease of spectral density of the signal components, varying within the corresponding
time intervals. At the same time, the non-linearity and non-locality of WV method cause
appearance of articial inter-modulation spectral components at combination frequencies
(artifacts) and may result also in suppression of weak spectral components of the signal by its
more intense or noisy parts (Cohen, 1989; Shkelev et al., 2002). To compensate the drawbacks
of SWF and WV data analysis methods, when they are used separately, and to keep their
strong features, the methods were combined in a proper way in the SWF-WV algorithm,
which uses various types of signal processing and ltration (with variable shape and size
of the analysis windows) in order to eliminate possible artifacts and to provide high spectral
and temporal resolution (Kislyakov et al., 2011; Shkelev et al., 2002).
To avoid the appearance of spurious spectra caused by the signal edge effects in the case
of analysis of real nite in time signal samples, the so called "weighting" functions with
smoothed edges are used. In the present study, the SWF spectrum S
k
of a discrete signal
151
s
n
consisting of N counts is calculated by a discrete Fourier transform (DFT) (Allen & Mills,
2004; Marple, 1986):
S
k
=
N1
n=0
wnd(n)s
n
exp
_
i
2nk
N
_
, k = 0, 1, ..., N 1 (6)
with a sliding window wnd(n). The following window functions are used (see also Pollock
(1999)):
wnd
1
(n) =
_
1 , 0 n < N
0 , n N
a rectangular window
wnd
2
(n) =
_
cos
2
_
n
2N
_
, 0 n < N/2
0 , n N/2
Hennings window
wnd
3
(n) =
1 6
_
2n
N
_
2
+6
_
2n
N
_
3
, 0 n < N/4
2
_
1
2n
N
_
3
, N/4 n < N/2
0 , n N/2
Parsens window.
(7)
The advantage of this method consists in its high performance speed, especially if the standard
algorithms of fast Fourier transform (FFT) are used in the calculations (Allen & Mills, 2004;
Marple, 1986). On the other hand, frequency resolution of SWF is reverse proportional to
the number of signal counts in the applied window. Therefore the size of window should be
sufciently large. This in its turn decreases the temporal resolution of the method. In practice,
the choice of particular type and width of the window is determined by dynamical features of
the analyzed signal.
An algorithm of the discrete WV transform is determined by the following expression:
P
mk
= P(mt, kf ) = 2t
2N1
n=0
_
z
m+n
z
mn
exp
_
i
nk
N
__
, {m, k} = 0, 1, 2, ..., 2N, (8)
where z
n
and z
n
are discrete values of the analyzed complex analytical signal made, as
determined above, of the real discrete signal and its Hilbert conjugate; t is a period of
discretization, and f = 1/(4Nt) is the frequency step. Note, that WV transform results
in real values only in the case of continuous functions integrated in innite limits (like in (5)).
The discrete WV transform (8) yields a complex function P
mk
, which is called as "pseudo-WV
transform" (Cohen, 1989). In that respect in practice only a module of P
mk
or its real part are
considered.
In course of the comparative study performed in Kislyakov et al. (2011); Shkelev et al. (2002)
it has been shown that WV data analysis technique enables higher spectral and temporal
resolution than that of the SWF. In Shkelev et al. (2002) the efciency of WV and SWF
methods was checked with various test signals, made as combinations of impulsive and
quasi-harmonic processes. By this, along with the meaningful spectrum, WV transform
generated specic articial spectral features (due to the non-linearity of the method). It has
been shown in that respect that superimposing of the higher resolution WVspectra with those
of lower resolution provided by SWF, may help to identify and to exclude these articial
spectral features form the consideration. Altogether, combined with each other the described
above SWF and WV methods provide an efcient data analysis algorithm characterized
by high sensitivity, high spectral and temporal resolution, and ability to detect complex
multi-signal modulations in the analyzed data records, enabling the dynamical spectra of
these modulations. For successful operation of the SWF-WV algorithm, the sampling cadence
of analyzed data series should provide sufcient number of the data points, e.g. 10, 000
points per realization. The length of the analyzed data series should be consistent with
the time scales of considered dynamic phenomena, i.e. the duration of an analyzed data
set should include at least several periods of the modulating oscillatory component. The
SWF-WV method appears the most efcient for the study of signals with non-stationary
complex modulations. In such cases the traditional Fourier transform and wavelet methods
are less efcient. This feature of the algorithm has been, in particular, used to distinguish
between the modulations, possibly caused by the large-scale transverse quasi-periodic motion
of the loops, which are the subject of the present study, and the modulations with frequency
drifts related with the electric current LCR-oscillations in the loops.
For the visualization of the whole variety of the detected modulations, so called averaged
spectral density plots are produced along with the dynamic spectra by the SWF-WV
algorithm. These plots are obtained by averaging of multiple instantaneous cuts of the
dynamic spectrum taken at given moments of time, so that short-living modulation features
also become clearly seen among the longer lasting modulation lines. These both types of
spectra (dynamic and averaged) enable the detection of the large-scale transverse oscillatory
dynamics of the radiating coronal magnetic loops.
The universality of SWF-WV algorithm resulted in its successful application in different
branches of space physics. The algorithmwas used for the diagnostics of intrinsic physical and
dynamical conditions in the stellar and planetary systems, solar/stellar winds, as well as in
solar and planetary radiation sources and planetary environments (Khodachenko et al., 2006;
Kislyakov et al., 2006; Panchenko et al., 2009; Zaitsev et al., 2003; 2004). Nowadays, the link
to SWF-WV data analysis algorithm is available for the scientic community via the on-line
catalogue of models and data analysis tools (http://europlanet-jra3.oeaw.ac.at/catalogue/),
developed within the JRA3-EMDAF (European modelling and data analysis facilities) activity
(http://europlanet-jra3.oeaw.ac.at/) of the European FP7 research infrastructure project
Europlanet-RI.
4. Diagnostics of large-scale oscillations of coronal loops by the analysis of VLF
modulations of microwave radiation
5. Instrumentation and modulations detection capabilities
We analyze the VLF (< 0.01 Hz) modulations of solar microwave bursts recorded in
Metshovi Radio Observatory (Finland) with the 14-mand 1.8 mradio telescope antenna at 37
GHz and 11.7 GHz, respectively. The key selection criterion for the analyzed microwave data
was their synchronism with the oscillating loops observed in extreme ultraviolet (EUV) by
TRACE (Aschwanden et al., 2002). The width of the antenna beam pattern of the Metshovi
radio telescope at 37 GHz is 2.4, the sensitivity of the receiver is about 0.1 sfu (10
23
W m
2
Hz
1
), and time resolution, 0.05 0.1 s. Therefore, at 37 GHz the spatial resolution of
the radio telescope is sufcient for identication of an active region that contains a radiating
source. This enables to analyze microwave radiation emitted directly from the region, imaged
in EUV (e.g., observed by TRACE), and to perform the comparison of the radiation features
153
and dynamics of coronal loops. At 11.7 GHz the radiation is collected from the whole solar
disk and the position of the radiating source cannot be resolved. However, even in this
case, by comparison with observations in other wavelengths and timing of the events, it is
usually possible to identify the microwave radiation features related to the energy release and
dynamic phenomena in particular active regions.
Variations of the background magnetic eld in a radiating source may cause not only the
amplitude modulation of microwave radio emission from solar active regions (according (1)),
but also could result in a certain frequency modulation (due to the dependence of electron
gyro-frequency
B
on the magnetic eld). The estimated width of the gyro-frequency variation
interval due to this effect is from several tens to several hundreds MHz. At the same time,
the bandwidth of the receiver at Metshovi is much larger than this interval and a possible
frequency modulation of the microwave radio emission cannot be resolved. Thus, one can
detect only the effects of varying magnetic eld, manifested in the intensity modulation of the
microwave signal, due to (1).
5.1 Observations and interpretation
In this subsection, we showhowthe analysis of long-periodic modulations of solar microwave
radiation accompanying the explosive events on the Sun may be used to obtain information on
the details of large-scale dynamics of the coronal loops associated with ares and the overall
structure of solar active regions. By this, the primary focus here is made on VLF modulations
of the radiation, emitted from active regions where TRACE observed in EUV at the same
time the large-scale oscillations of coronal loops. Careful check of the solar microwave radio
emission records available at Metshovi revealed several events which coincide in time with
the EUV observations of oscillating coronal loops. Below these events are considered in
details.
Figure 3c shows the intensity prole and dynamic spectrum of VLF modulations of
microwave emission from the active region AR8910 on the limb (see Figure 3a,b) where a
group of oscillating loops (Figure 4a) was observed by TRACE after M2.0 are on 2000-Mar-23,
at 11:30-12:00 UT (Aschwanden et al., 2002). These observations were performed at 37 GHz,
and the spatial resolution of the Metshovi radio telescope was sufcient to resolve the
radiating source in the active region AR8910.
A remarkable feature of the VLF modulation dynamic spectrum in Fig. 3c,d and the averaged
spectral density plot in Fig. 3d consists in the presence of several "modulation pairs". These
are the modulations (a) at 1.7 mHz ( 10 min) and 3.4 mHz ( 4.9 min); (b) at 6.0 mHz ( 2.8
min) and 12.0 mHz ( 1.4 min), as well as (c) at 7.8 mHz ( 2.1 min) and 15.6 mHz ( 64
s). According to the considerations in Section 2, these "modulation pairs" could indicate the
transverse oscillating loops with the periods, corresponding to the main frequencies of the
pairs, i.e. 10 min; 2.8 min, and 2.1 min for the cases (a), (b), and (c), respectively. By
this, the rst "modulation pair" (case (a)) as a signature of the loop transverse oscillation with
a period 10, ts quite well the results of TRACE observations, which found the oscillating
loop with approximately the same period (615 s) (Aschwanden et al., 2002; Schrijver et al.,
2002). As it can be seen from the dynamical spectrum in Figure 3c, the spectral resolution of
the analysis performed in this particular case was about 0.3 mHz. By this, the frequency 1.62
mHz corresponding to the detected TRACE period of 615 s is denitely within the frequency
c
F
r
e
q
u
e
n
c
y
,
H
z
I
n
t
e
n
s
i
t
y
,
a
.
u
0.01
0.008
0.006
0.004
0.002
0
1E+4
8E+3
4E+3
11:20 11:40 12:00
Time (UT)
d
R
e
l
a
t
i
v
e
s
p
e
c
t
r
a
l
d
e
n
s
i
t
y
Burst on 23.03.2000
Spectrum averaged on T=90-240
at 37 Ghz
30
25
20
15
10
5
0
0 4 6 8 10 12 14 16
Frequency, mHz
2
1.7 mHz
3.4 mHz
6.0 mHz
7.8 mHz
8.4 mHz
12.0 mHz
15.6 mHz
Solar burst on 23.03.2000
at 37 GHz
Dynamic
spectrum
burst on
2000-Mar-23
a b
S
O
H
O
/
M
D
I
M
a
g
n
e
t
o
g
r
a
m
,
2
0
0
0
-
M
a
r
-
2
3
S
O
H
O
/
E
I
T
3
0
4
,
2
0
0
0
-
M
a
r
-
2
3
Fig. 3. (a) SOHO/MDI Magnetogram of the Sun on 2000-Mar-23, white arrow points at the
active region AR8910; (b) The Sun image in 304 on 2000-Mar-23 from SOHO/EIT, white
arrow points at the active region AR8910; (c) Intensity prole and corresponding VLF
modulation dynamic spectrum of the microwave radiation, recorded from the active region
AR8910 on 2000-Mar-23, at 11:30-12:00; Color codes the dynamic spectral relative intensity
(arbitrary units), more dark features correspond to stronger (better pronounced)
modulations; (d) averaged spectral density of the VLF modulation.
interval 1.7 0.3 mHz of the modulation feature revealed by the analysis of the microwave
radiation.
155
Fig. 4. (a) Transverse oscillating coronal loops observed by TRACE in the active region
AR8910 after an M2.0 are on 2000-Mar-23 at 11:30-12:00 UT (Aschwanden et al., 2002); (b)
Phase comparison of the 10 min modulation component of the microwave emission on
2000-Mar-23 and the amplitude of the corresponding 615 s oscillation of the TRACE loop.
A higher level of the second harmonic in the modulation pair 1.7 mHz ( 10 min) and
3.4 mHz ( 4.9 min) is very likely due to the fact that in this particular case two different
mechanisms, modulating the microwave emission of the loop, are by chance manifested
simultaneously. The rst mechanism is that considered in this paper, which is connected
with a large-scale transverse oscillation of the loop. The second mechanism is due to
the parametric resonance between 5-min velocity oscillations in the solar photosphere and
acoustic oscillations of coronal magnetic loop modulating the microwave emission (Zaitsev
et al., 2008). The effect consists in simultaneous excitation in the loop, which occasionally
appeared to have a resonant frequency close to 10 min, of oscillations with periods 5 min,
10 min, and 3 min, which correspond to the 5-min pumping frequency of the photospheric
convection velocity oscillations, subharmonic, and the rst upper frequency of the parametric
resonance, respectively (Zaitsev et al., 2008; Zaitsev & Kislyakov, 2006).
It makes no sense to search in TRACE data for the signatures of other oscillating loops
(cases (b) and (c)), indicated by the VLF modulations of the microwave radiation during
the 2000-Mar-23 event, since with the usual 40 s image sampling cadence of TRACE and
the 4-point resolution limit of the instrument (Aschwanden et al., 2002) the fastest detectable
by TRACE period is about 3 min. The remaining short-periodic "non-paired" modulation
feature at 8.4 mHz ( 1.9 min) may also be a part of a "modulation pair", of which the
second harmonic counterpart could not be resolved in the VLF spectrum due to the strong
contamination of the analyzed microwave signal. Such weak higher harmonic components
may be as well a signature of another oscillatory process, which is unrelated to the large-scale
transverse motion of loops, e.g., a sausage-type MHD wave excited in a loop. Detailed
analysis of this special case remains however beyond the scope of the present study.
An additional conrmation of the fact that 10 min modulation of the microwave radiation
emitted from the active region AR8910 on 2000-Mar-23, is connected with the transverse
oscillatory motion of a coronal loop, shown in Fig. 4a, and that it is in very likely related to the
motion of the emission diagram pattern, comes from the graphs in Fig. 4b. This gure enables
phase comparison for the transverse motion of the loop (observed with TRACE Aschwanden
et al. (2002)) and the ltered 1.7 mHz ( 10 min) component of the radiation. The last
characterizes temporal behaviour of the radio emission received from the oscillating loop.
The shifted phase ( ) means that the maxima and minima of the radiation part controlled
by the loop motion correspond to the specic orientations of the loop and are connected with
a certain direction of the emission diagram relative to the observer.
The microwave burst on 2001-Sep-07 represents another example of manifestation of the
coronal loop transverse oscillations in modulation of solar radio emissions. The burst was
produced during M-are activity at 15:30 UT in the active region AR9601, close to the solar
disc center (see Figure 5a,b), where TRACE observed a group of oscillating loops, immediately
after the are (Aschwanden et al., 2002). The corresponding microwave radiation record was
made at 11.7 GHz with the Metshovi radio telescope. At this frequency, the Metshovi
antenna cannot resolve the position of a radiating source, and the emission from the whole
solar disk contributed to the analyzed microwave intensity prole. At the same time, as can
be seen in Fig. 5c, which presents the analyzed microwave radiation record with the burst
and its VLF modulation dynamic spectrum, the spectral features related to the processes in
the aring active region AR9601 can easily be identied by timing of the event. In particular,
the dynamic spectrum of VLF modulations of the microwave radiation exhibits several lines,
which start simultaneously with the impulsive phase of the are (at 15:30 UT). These may be
the signatures of different oscillating loops excited by the are. By this, most of the oscillations
(e.g. the dynamic spectrum lines) decay at the time intervals > 20 min. Unfortunately for
this particular event it is impossible to determine exact duration of each of these decaying
modulations because the available microwave radiation record does not cover the end of the
event. As it can be seen in Fig. 5c, some of the dynamic spectrum lines continue beyond the
analyzed record time frame.
Three "modulation pairs" can be identied in the dynamic and averaged spectra in Fig. 5c,d:
(a) 1.8 mHz ( 9.2 min) and 3.6 mHz ( 4.6 min); (b) 2.7 mHz ( 6.2 min) and 5.4 mHz
( 3.1 min); as well as (c) 4.3 mHz ( 3.8 min) and 8.6 mHz ( 1.9 min), which may be
the signatures of transverse oscillating loops with periods 9.2 min, 6.2 min, and 3.8
min, respectively. We note that the loop periods in the cases (a) and (b) are consistent with
the 6-10 min oscillating loops observed with TRACE (Aschwanden et al., 2002), whereas the
shorter period oscillation (case (c)) cannot be resolved by TRACE because of the relatively
long image sampling cadence. The modulation at 5.4 mHz ( 3.1 min) may also be a weak
third harmonic produced by the 9.2 min oscillating loop. If this is true, then the line at 2.7 mHz
( 6.2 min) will have no a pair-companion, and one should exclude the possibility of the 6.2
min transverse oscillating loop. A "non-paired" weak modulation feature at 6.4 mHz ( 2.6
min) may be a signature of an oscillating loop with not resolved second harmonic. At the
same time, as for the 2000-Mar-23 event, weak short-period harmonics may be the signatures
of oscillatory processes that are unrelated to the transverse motion of a loop, but caused only
by a changing magnetic eld in the radiating source.
The strong modulation line at 0.6 mHz ( 27.7 min) should be considered separately from all
other modulations mentioned above. The dynamical spectrum in Fig. 5c, as well as a separate
study of VLF modulations of the microwave radiation recorded before the faring burst at 15:30
UT, reveal the presence of the 27.7 min component also before the are. In view of the fact
157
Fig. 5. (a) SOHO/MDI Magnetogram of the Sun on 2001-Sep-07, white arrow points at the
active region AR9601; (b) The Sun image in 304 on 2001-Sep-07 from SOHO/EIT, white
modulation dynamic spectrum of the microwave burst on 2001-Sep-07, at 15:30-15:50
associated with an M-are in the active region AR9601; Color codes the dynamic spectral
relative intensity (arbitrary units), more dark features correspond to stronger (better
pronounced) modulations; (d) averaged spectral density of the VLF modulation.
that the analyzed microwave emission (at 11.7 GHz) was received from the whole solar disk,
the 27.7 min modulated part of radiation very likely originates in another active region. It
may also be connected with a kind of global solar seismology process.
The last example of possible manifestation of transverse oscillations of coronal loops in
microwaves which we present here, is the event on 2001-Sep-15, when TRACE observed
oscillating loops, associated with M are at 11:23 UT in the active region AR9608, close to
the limb (see Figure 6a,b). Similar to the case of the microwave burst on 2001-Sep-07, the
event on 2001-Sep-15 was observed at 11.7 GHz, and the microwave radiation source in
AR9608 was not resolved by the Metshovi antenna. Thus, the emission from the whole
solar disk contributed to the analyzed microwave intensity prole. However, as for the
burst on 2001-Sep-07, all the spectral features related to the aring active region AR9608
can be identied by event timing. As it can be seen in Fig. 6c, the dynamic spectrum
of VLF modulations of the microwave radiation consists of several lines, most of which
begin simultaneously with the impulsive phase of the are at 11:23 UT. These lines may be
associated with oscillatory processes in the active region loops triggered by the are. All the
post-are oscillations decay at the time intervals from 20 min up to 1 hour.
Fig. 6d shows the averaged spectral density of the VLF modulations of the microwave burst
on 2001-Sep-15. At least one "modulation pair": 1.3 mHz ( 12.8 min) and 2.6 mHz ( 6.4
min) can be identied among the detected modulation lines. It is very likely connected
with a transverse oscillating loop having the period 12.8 min. This result agrees with
the reported TRACE observations of the oscillating loop in the active region AR9608 with a
period 12 15 min. Other detected in the microwave record on 2001-Sep-15 (see Fig. 6c,d),
more short-periodic modulations at 3.8 mHz ( 4.4 min) and 5.2 mHz ( 3.2 min) could
be higher-order harmonics produced by the 12.8 min oscillating loop, or the modulations
associated with oscillatory processes not connected with the transverse motion of loops. They
may also be the signatures of oscillatory processes in small loops, which cannot be seen in
the TRACE EUV movies due to the limitations of the operated observing mode. Additionally
should be mentioned a relatively weak line at 0.7 mHz ( 23.8 min). Its possible second
harmonic could contribute to the broad line near 1.3 mHz, which is identied as a main
frequency line in another modulation pair. If that is the case, then this may be a signature of
another transverse oscillating loop with the period 23.8 min. Verwichte et al. (2010) reported
recently detection of such an oscillating loop related with the considered aring event in the
the active region AR9608.
A special remark deserves also the long lasting ultra-low-frequency (ULF) modulation
at 0.3 mHz ( 56 min), clearly visible in both, the dynamic and averaged, spectra of
the long-periodic modulations of solar microwave radiation on 2001-Sep-15 in Figs. 6c,d.
Similarly to the 27 min line in the case of the 2001-Sep-07 burst, this ULF modulation appears
before the aring burst and lasts much longer than all other modulation lines in the spectrum
(Fig. 6c). Therefore, it cannot be related to the are in the active region AR9608 and consequent
post-are dynamics of coronal loops. This modulation feature is probably connected with the
solar seismology processes, or a slow dynamics of another active region. As an additional
argument in support of the solar global i.e., helioseismic nature of the ULF (
0
< 0.6 mHz,
i.e. > 30 min) modulations may be the fact that these modulations are usually detected
in the radiation records made at 11.7 GHz when the radio emission from the whole solar
disk contributes to the analyzed microwave intensity prole. The radiation emitted from
the spatially resolved separate active regions, for example, at 37 GHz with Metshovi radio
telescope, does not exhibit any ULF modulation features. In more details the ULF modulations
of solar microwave radiation are considered in Kislyakova et al. (2011).
159
c d
R
e
l
a
t
i
v
e
s
p
e
c
t
r
a
l
d
e
n
s
i
t
y
Burst on 15.09.2001
Spectrum averaged on ~ 40 min
at 11.7 Ghz,
0.0025
0.002
0.0015
0.001
0.0005
0
0 2 3 4 5 6 7
Frequency, mHz
1
3.8 mHz
2.6 mHz
0.3 mHz
1.3 mHz
5.2 mHz
F
r
e
q
u
e
n
c
y
,
m
H
z
I
n
t
e
n
s
i
t
y
,
a
.
u
10
8
6
4
2
0
Dynamic spectrum,
burst on 2001-Sep-15
8
7.5
7
10:00 11:00 12:00
Time (UT)
6.5
6
5.5
14:00 13:00
Solar burst on 15.09.2001
at 11.7 Ghz
a b
S
O
H
O
/
M
D
I
M
a
g
n
e
t
o
g
r
a
m
,
2
0
0
1
-
S
e
p
-
1
5
S
O
H
O
/
E
I
T
3
0
4
,
2
0
0
1
-
S
e
p
-
1
5
0.7 mHz
Fig. 6. (a) SOHO/MDI Magnetogram of the Sun on 2001-Sep-15, white arrow points at the
active region AR9608; (b) The Sun image in 304 on 2001-Sep-15 from SOHO/EIT, white
modulation dynamic spectrum of the microwave burst on 2001-Sep-15, at 11:23-12:15
associated with an M-are in the active region AR9608; Color codes the dynamic spectral
relative intensity (arbitrary units), more dark features correspond to stronger (better
pronounced) modulations; (d) averaged spectral density of the VLF modulation.
5.2 On the magnetic eld variations, estimated from VLF spectra
It is impossible, using the available data, to perform an exact calculation for the amplitude of
magnetic eld variations. That is because the analyzed microwave signals were recorded
in relative units without calibration to the radiation intensity scale. At the same time,
taking into account known values of the maximal intensity measured during the radio
bursts and the lower limit of sensitivity of the Metshovi receiver, the intensity modulation
amplitude I
can be estimated roughly from the obtained VLF spectra, by their comparison
with the spectra of specially created modelling signals (Khodachenko et al., 2005). For the
considered in the paper events this estimation gives the value of the relative variation of
intensity I
/I
0
10
2
10
1
. Assuming, as an upper rough limit, that this variation of
intensity is connected only with the varying magnetic eld, we come to a relation I
/I
0
=
(B/B
0
)
1 = ((B/B
0
+1)
1), where B/B

0
is the relative variation of magnetic eld,
and = 0.22 +0.9 is the power index of magnetic eld in the equation (1). For 2 < < 7
one gets that = 1.58 6.08. For the case of small relative variations of the magnetic eld we
nally obtain I
/I
0
B/B
0
= 10
2
10
1
, which for B
0
= 100G gives the amplitude
B = (0.16 6.3) G.
6. Discussion, conclusions and perspectives
The analysis of VLF modulations of solar microwave bursts presented in this work shows
good temporal coincidence of the modulations and their oscillatory parameters with the
observed decaying large-scale transverse oscillations of the coronal loops triggered by ares.
This indicates about a physical link between the oscillatory motion of the loops and variations
of the observed radio emission. As a working hypothesis to take this link into account, a loop
with the propagating beams of non-thermal particles which produce microwave emission
due to the electron gyro-synchrotron mechanism, has been considered. As pointed out by
Schrijver et al. (2002), who considered several cases of transverse oscillations of coronal loops
observed with TRACE (including the event on 2000-Mar-23 addressed in the present paper),
in almost all cases the oscillating loops lie at, or near, the large-scale separatrices, or near the
sites involved in reconnection. These regions may be the sources of non-thermal particles
injected into the loops and generating microwave emission there. Moreover, in the case on
2000-Mar-23 the loop oscillation happened in response to a aring event located at the loop
base Schrijver et al. (2002). That could provide a direct input of energetic particles into the
loop.
In the most general case, thermal bremsstrahlung mechanism of the radiation should be also
considered, besides of the gyro-synchrotron, for the analyzed frequency range of the solar
microwave emission. If the last mechanism assumes that there are high energy electrons
passing through the magnetic loop, the rst one is connected with the radiation of hot plasma
heated by the electron beams in the chromospheric footpoints of the loop. A comparative
study of contribution of the bremsstrahlung and gyro-synchrotron radiation to the microwave
emission of a are, performed in Urpo et al. (1994), shows that thermal bremsstrahlung
is more important for the microwave events that have an intensity of the order of or less
than 100 SFU, with the exception of cases when the electron spectrum is sufciently hard.
Therefore, correct interpretation of the microwave radiation source requires consideration
of both mechanisms, for example by involving of the hybrid thermal/nonthermal model
of the solar are emission (Holman & Benka, 1992). However, looking at a possibility of
an oscillatory behaviour of the microwave radiation source which constitutes the primary
subject of the present study, we notice that in the case of the bremsstrahlung mechanism
it is possible only for a varying energy deposition into the system, i.e. a varying ow of
non-thermal particles heating the loop footpoints. In view of the unclearness of how the
post-are transverse oscillating loop may modulate the source of accelerated electrons, we
built our analysis with the assumption that the non-thermal particle population remains
161
more-or-less stable, and all the variations of the observed microwave radiation are connected
with the non-thermal gyro-synchrotron part of the radiation, modulated by the large-scale
oscillatory motion of the loop. The bremsstrahlung component, even being present in the
microwave radiation, does not contribute to the analyzed oscillating part of the emission
provided by the gyro-synchrotron mechanism.
Our analysis here is based on the assumption of an optically thin microwave source. In this
case the radiation intensity is proportional to emissivity
, given by equation (Dulk, 1985)
3.3 10
24
10
0.52
NB(sin )
0.43+0.65
_

B
_
1.220.9
, (9)
where is radiation frequency, N is the number of electrons per cubic centimeter with energy
higher than 10 keV, B is magnetic eld,
B
= eB/(2m
e
c), is the electron-cyclotron frequency,
and is the electron energy spectrum index. This fact has been used for obtaining our basic
equation (1) in Section 1. In the opposite case of an optically thick source, the intensity of
microwave radiation is proportional to the effective temperature T
e f f
, i.e., to the ratio of
emissivity
and absorption coefcient
. For the last, Dulk (1985) provides the following

expression:
1.4 10
9
10
0.22
(N/B)(sin )
0.09+0.72
_

B
_
1.30.98
. (10)
Therefore, in the case of an optically thick source the dependence of radiation intensity on the
varying magnetic eld B(t) and direction to the observer (t) will be different than that given
by the equation (1). However, the character of this dependence, i.e. I
(sin((t)))
k
B(t)
l
,
where k and l are the numbers depending on the non-thermal electron spectral index , will
remain the same. Therefore, one may expect similar manifestation of the varying B(t) and
(t) in the modulation of the received microwave emission also in the case of an optically
thick source.
Analysis of LF and VLF modulations of solar microwave radiation is a relatively new
direction in solar radio astronomy which appears nowadays a subject of a certain interest.
VLF variations of solar microwave radiation intensity may be related to slow variations of
magnetic eld in a radiating source, as well as to large-scale motions of coronal structures
containing the radiating source. Joint action of two radiation modulating factors: (i) the
quasi-periodic uctuation of magnetic eld and (ii) motion of the radio emission diagram,
in the case of a transverse oscillating coronal loop results in essentially non-sinusoidal
(non-harmonic) character of the signal received by a remote observer, with strongly
pronounced two rst harmonics (at the main and double frequency of the oscillation),
called here as "modulation pairs". Such specics of the VLF spectrum has been used for
the identication of transverse oscillating loops triggered by ares. The analysis of solar
microwave records has been performed with an algorithm based on Sliding Window Fourier
transform and Wigner-Ville techniques. This high sensitive algorithm provides, signicant
spectral and temporal resolution which enable clear detection of VLF "modulation pairs" in
the solar aring microwave radiation (microwave bursts). Comparison of parameters of these
"modulation pairs" with the simultaneous TRACE observations in EUV of the corresponding
solar active regions enabled to associate some of the paired VLF modulation features with the
large-scale transverse oscillations of coronal loops. The "non-paired" features also detected
in the VLF modulation dynamic spectra, may be either parts of "modulation pairs" in which
the second harmonic cannot be resolved because of the strong contamination of the analyzed
signal, or be the signatures of other oscillatory processes (MHDmodes) in the loops, unrelated
to their large-scale transverse motion, e.g., sausage-type MHD waves.
The presence of "modulation pairs" in the VLF spectra of solar microwave radiation is
considered here as an indication of the transverse oscillating coronal loops. However, the
exact characterization of the transverse motion of radiating loops by the dynamical spectra
of microwave emission needs a quantitative analysis of the measured radio signal and
superimposing these results with a precise calculation of the radiation from the loop, taking
into account the loop position relative observer. This topic requires a dedicated study. It
outlines the general direction for further development of the ideas expressed in the paper.
Besides of that, of certain interest appear the long lasting ULF modulations of solar microwave
radiation detected at < 0.6 mHz (> 30 min) in the absence of bursts, i.e. during the periods
of quiet Sun. It is remarkable that these modulations are not visible in the emissions from
separate active regions, recorded in particular at 37 GHz with Metshovi radio telescope. But
they appear in the integrated radiation received from the whole solar disk at 11.7 GHz. This
fact may be considered as an argument in support of the global helioseismic nature of these
ULF modulations, which also require further more detailed study.
7. Acknowledgements
This work was supported by the Austrian Fond zur Frderung der wissenschaftlichen
Forschung (project P21197-N16). The authors are thankful to the JRA3/EMDAF
(European Modelling and Data Analysis Facility; http://europlanet-jra3.oeaw.ac.at) a
subdivision of the European research infrastructure EUROPLANET-RI and EU FP7 project
IMPEx (http://impex-fp7.oeaw.ac.at) for the support of their research communication and
collaboration visits. The authors would like to thank M.Aschwanden for useful discussions
and the data regarding the transverse large-scale dynamics of coronal loops.
8. References
Allen, R.L. & Mills, D.W. (2004). Signal analysis: time, frequency, scale, and structure. IEEE Press,
Wiley-Interscience, ISBN:0-471-23441-9.
Aschwanden, M.J.; Fletcher, L.; Schrijver, C.J.; Alexander, D. (1999). Coronal Loop Oscillations
Observed with the Transition Region and Coronal Explorer. The Astrophysical Journal,
Vol. 520, No. 2, 880894, (DOI:10.1086/307502).
Aschwanden, M.J. (2002). Particle acceleration and kinematics in solar ares - A Synthesis
of Recent Observations and Theoretical Concepts (Invited Review). Space Science
Review, Vol. 101, No. 1, 1227, (DOI:10.1023/A:1019712124366).
Aschwanden, M.J.; DePontieu, B.; Schrijver, C.J.; Title, A. (2002). Transverse Oscillations in
Coronal Loops Observed with TRACE II. Measurements of Geometric and Physical
Parameters. Solar Physics, Vol. 206, No. 1, 99132, (DOI:10.1023/A:1014916701283).
Cargill, P.J.; Goodrich, C.C.; Vlahos, L. (1988). Collisionless shock formation and the prompt
acceleration of solar are ions. Astronomy and Astrophysics, Vol. 189, No. 1-2, Jan.,
254262, ISSN 0004-6361.
Cohen, L. (1989). Time-frequency distributions - A review. IEEE Proc., Vol. 77, July-1989,
941981, ISSN 0018-9219.
163
Dulk, G.A. (1985). Radio emission from the sun and stars. IN: Annual review of astronomy and
astrophysics, Vol. 23(A86-14507 04-90), 169224, Palo Alto, CA, Annual Reviews, Inc.,
(DOI:10.1146/annurev.aa.23.090185.001125).
Dulk, G.A. & Marsh, K.A. (1982). Simplied expressions for the gyrosynchrotron radiation
from mildly relativistic, nonthermal and thermal electrons. The Astrophysical Journal,
Vol. 259, Aug-1982, 350358, (DOI:10.1086/160171).
Hardy, S.J.; Melrose, D.B.; Hudson, H.S. (1998). Observational tests of a double loop model for
solar ares. Publications Astronomical Society of Australia, Vol. 15, No. 3, 318324.
Holman, G.D. & Benka, S.G. (1992). A hybrid thermal/nonthermal model for the energetic
emissions from solar ares. The Astrophysical Journal, Vol. 400, No. 2, L79L82, ISSN
0004-637X, (DOI:10.1086/186654).
Holman, G.D. & Pesses, M.E. (1983). Solar type II radio emission and the shock drift
acceleration of electrons. The Astrophysical Journal, Vol. 267, No. 15, 837843, ISSN
0004-637X, (DOI:10.1086/160918).
Karlick, M. & Kosugi, T. (2004). Acceleration and heating processes in a
collapsing magnetic trap. Astronomy and Astrophysics, V. 419, 11591168,
(DOI:10.1051/0004-6361:20034323).
Khodachenko, M.L.; Haerendel, G.; Rucker, H.O. (2003). Inductive electromagnetic effects in
solar current-carrying magnetic loops. Astronomy and Astrophysics, Vol. 401, 721732,
(DOI:10.1051/0004-6361:20030146).
Khodachenko, M.L.; Zaitsev, V.V.; Kislaykov, A.G.; Rucker, H.O.; Urpo, S. (2005).
Low-frequency modulations in the solar microwave radiation as a possible indicator
of inductive interaction of coronal magnetic loops. Astronomy and Astrophysics, Vol.
433, No. 2, 691699, (DOI:10.1051/0004-6361:20041988).
Khodachenko, M.L.; Rucker, H.O.; Kislaykov, A.G.; Zaitsev, V.V.; Urpo, S. (2006).
Microwave Diagnostics of Dynamic Processes and Oscillations in Groups of
Solar Coronal Magnetic Loops. Space Science Reviews, Vol. 122, No. 1-4, 137148,
(DOI:10.1007/s11214-006-7767-0).
Khodachenko, M.L.; Zaitsev, V.V.; Kislyakov, A.G.; Stepanov, A.V. Equivalent
Electric Circuit Models of Coronal Magnetic Loops and Related Oscillatory
Phenomena on the Sun. (2009). Space Science Reviews, Vol. 149, No. 1-4, 83117,
(DOI:10.1007/s11214-009-9538-1).
Khodachenko, M.L.; Kislyakova, K.; Zaqarashvili, T.V.; Kislyakov, A.G.; Panchenko,
M.; Zaitsev, V.V.; Arkhypov, O.V.; Rucker, H.O. (2011). Possible
manifestation of large-scale transverse oscillations of coronal loops in solar
microwave emission. Astronomy and Astrophysics, Vol. 525, CiteID A105,
(DOI:10.1051/0004-6361/201014860).
Kislyakov, A. G.; Zaitsev, V. V.; Stepanov, A. V.; Urpo, S. (2006). On the Possible Connection
between Photospheric 5-Min Oscillation and Solar Flare Microwave Emission. Solar
Physics, Vol. 233, No. 1, 89106, (DOI:10.1007/s11207-006-2850-y).
Kislyakova, K.G.; Zaitsev, V.V.; Urpo, S.; Riehokainen, A. (2011). Long-period oscillations
of the solar microwave emission. Astronomy Reports, Vol. 55, No. 3, 275283,
(DOI:10.1134/S1063772911030036).
Kislyakov, A.G.; Shkelev, E.I.; Lupov, S.Y.; Kislyakova, K.G. (2011). Parameters
of astrophysical objects according to their electromagnetic emission intensity
modulation data. I. Observational data processing algorithms. Bulletin of the Nizhny
Novgorod State University, No.2(1), 4654, (in russian).
LaRosa, T.N. &Moore, R.L. (1993). AMechanismfor Bulk Energization in the Impulsive Phase
of Solar Flares: MHD Turbulent Cascade. The Astrophysical Journal, Vol. 418, 912918,
(DOI:10.1086/173448).
Leka, K.D.; Caneld, R.C.; McClymont, A.N.; van Driel-Gesztelyi, L. (1996). Evidence
for Current-carrying Emerging Flux. The Astrophysical Journal, Vol. 462, 547560,
(DOI:10.1086/177171).
Marple, S. L., Jr. (1986). Digital spectral analysis with applications. Prentice-Hall, Inc. Upper
Saddle River, NJ, USA, ISBN:0-132-14149-3.
Miller, J.A.; LaRosa, T.N.; Moore, R.L. (1996). Stochastic Electron Acceleration by Cascading
Fast Mode Waves in Impulsive Solar Flares. The Astrophysical Journal, Vol. 461,
445464, (DOI:10.1086/177072).
Miller, J.A.; Cargill, P.J.; Emslie, A.G.; Holman, G.D.; Dennis, B.R.; LaRosa, T.N.; Winglee,
R.M.; Benka, S.G.; Tsuneta, S. (1997). Critical issues for understanding particle
acceleration in impulsive solar ares. Journal of Geophysical Research, Vol. 102, No.
A7, 1463114660, (DOI:10.1029/97JA00976).
Moreton, G.E. & Severny, A.B. (1968). Magnetic Fields and Flares in the Region CMP 20
September 1963. Solar Physics, Vol. 3, No. 2, 282297, (DOI:10.1007/BF00155163).
Nakariakov, V. M.; Foullon, C.; Verwichte, E.; Young, N. P. (2006). Quasi-periodic
modulation of solar and stellar aring emission by magnetohydrodynamic
oscillations in a nearby loop. Astronomy and Astrophysics, Vol. 452, No. 1, 343346,
(DOI:10.1051/0004-6361:20054608).
Panchenko, M.; Khodachenko, M. L.; Kislyakov, A. G.; Rucker, H. O.; Hanasz, J.; Kaiser, M.
L.; Bale, S. D.; Lamy, L.; Cecconi, B.; Zarka, P.; Goetz, K. (2009). Daily variations of
auroral kilometric radiation observed by STEREO, Geophysical Research Letters, Vol.
36, No. 6, CiteID L06102, (DOI:10.1029/2008GL037042).
Pollock, D.S.G. (1999). A handbook of time-series analysis, signal processing and dynamics. London:
Academic Press, ISBN:0-12-560990-6.
Schrijver, C. J.; Title, A. M.; Berger, T. E.; Fletcher, L.; Hurlburt, N. E.; Nightingale, R. W.; Shine,
R. A.; Tarbell, T. D.; Wolfson, J.; Golub, L.; Bookbinder, J. A.; Deluca, E. E.; McMullen,
R. A.; Warren, H. P.; Kankelborg, C. C.; Handy, B. N.; de Pontieu, B. (1999). A new
view of the solar outer atmosphere by the Transition Region and Coronal Explorer.
Solar Physics, Vol. 187, No. 2, 261302, (DOI:10.1023/A:1005194519642).
Schrijver, C.J.; Aschwanden, M.J.; Title, A.M. (2002). Transverse oscillations in coronal loops
observed with TRACE I. An Overview of Events, Movies, and a Discussion of
Common Properties and Required Conditions. Solar Physics, Vol. 206, No. 1, 6998,
(DOI:10.1023/A:1014957715396).
Schrijver, C.J. & Brown, D.S. (2000). Oscillations in the Magnetic Field of the Solar Corona in
Response to Flares near the Photosphere. The Astrophysical Journal, Vol. 537, No. 1,
L69L72, (DOI:10.1086/312753).
Shkelev, E.I.; Kislyakov, A.G.; Lupov, S.Y. (2002). Methods of decreasing of a cross-modulation
effects in the Wigner-Ville distribution. Izv. Vyss. Uchebn. Zaved., Ser. Radioz. (transl.
as Radiophys. Quantum Electron.), Vol. 45, No. 5, 433-442.
Spangler, S.R. (2007). A Technique for Measuring Electrical Currents in the Solar Corona. The
Astrophysical Journal, Vol. 670, No. 1, 841848, (DOI:10.1086/521995).
Tan, B.; Ji, H.; Huang, G.; Zhou, T.; Song, Q.; Huang, Y. (2006). Evolution of Electric Currents
Associated with Two M-Class Flares. Solar Physics, Vol. 239, No. 1-2, 137148,
(DOI:10.1007/s11207-006-0120-7).
165
Urpo, S.; Bakhareva, N.M.; Zaitsev, V.V.; Stepanov, A.V. (1994). Comparison of mm-wave
and X-ray diagnostics of are plasma. Solar Physics, Vol. 154, No. 2, 317334, ISSN
0038-0938, (DOI:10.1007/BF00681102).
Verwichte, E.; Foullon, C.; Van Doorsselaere, T. (2010). Spatial Seismology of a Large Coronal
Loop Arcade from TRACE and EIT Observations of its Transverse Oscillations. The
Astrophysical Journal, Vol. 717, No. 1, 458467, (DOI:10.1088/0004-637X/717/1/458).
Ville, J. (1948). Cbles et Transm., Vol. 2A, 61.
Wigner, E. (1932). On the Quantum Correction For Thermodynamic Equilibrium. Phys. Rev.,
Vol. 40, No. 5, 749759, (DOI:10.1103/PhysRev.40.749).
Zaitsev, V.V. & Stepanov, A.V. (1992). Towards the circuit theory of solar ares. Solar Physics,
Vol. 139, No. 2, 343356, ISSN 0038-0938, (DOI:10.1007/BF00159158).
Zaitsev, V.V.; Stepanov, A.V.; Urpo, S.; Pohjolainen, S. (1998). LRC-circuit analog of
current-carrying magnetic loop: diagnostics of electric parameters. Astronomy and
Astrophysics, Vol. 337, 887896.
Zaitsev, V.V.; Kislyakov, A.G.; Urpo, S.; Shkelev E.I. (2001a). Observational evidence for
energy accumulation and dissipation in coronal magnetic loops. Izv. Vyssh. Uchebn.
Zaved., Ser. Radioz. (transl. as Radiophys. & Quant. Electronics), Vol. 44, No. 9, 697709,
(DOI:0033-8443/01/4409-0697$25.00).
Zaitsev, V.V.; Kislyakov, A.G.; Stepanov, A.V.; Urpo, S.; Shkelev E.I. (2001b).
Low-frequency pulsations of coronal magnetic loops. Izv. Vyssh. Uchebn. Zaved.,
Ser. Radioz. (transl. as Radiophys. & Quant. Electronics), Vol. 44, No. 1-2, 3652,
(DOI:0033-8443/01/441-2-0036$25.00).
Zaitsev, V.V.; Kislyakov, A.G.; Urpo, S.; Stepanov, A.V.; Shkelev E.I. (2003). Spectral-Temporal
Evolution of Low-Frequency Pulsations in the Microwave Radiation of Solar Flares.
Astronomy Reports, Vol. 47, No. 10, 873882, (DOI:10.1134/1.1618999).
Zaitsev, V.V.; Kislyakov, A.G.; Stepanov, A.V.; Kliem, B.; Frst, E. (2004). Pulsating
Microwave Emission from the Star AD Leo. Astronomy Letters, Vol. 30, 319324,
(DOI:10.1134/1.1738154).
Zaitsev, V. V.; Kislyakov, A. G.; Kislyakova, K. G. (2008). Parametric resonance in the solar
corona. Cosmic Research, Vol. 46, No. 4, 301308, (DOI:10.1134/S0010952508040035).
Zaitsev, V. V. & Kislyakov, A. G. (2006). Parametric excitation of acoustic oscillations
in closed coronal magnetic loops. Astronomy Reports, Vol. 50, No. 10, 823833,
(DOI:10.1134/S1063772906100076).
Part 2
Medical Applications

6
Spectral Analysis of
Heart Rate Variability in Women
Ester da Silva
1,2
, Ana Cristina S. Rebelo
1
, Nayara Y. Tamburs
2
,
Mariana R. Salviati
2
, Marcio Clementino S. Santos
3
and Roberta S. Zuttin
2
1
Federal University of So Carlos, So Carlos, SP,
2
College of Health Sciences, Methodist University of Piracicaba, Piracicaba, SP,
3
Par State University, Belm, PA,
Brazil
1. Introduction
This chapter discusses heart rate variability (HRV) to understand autonomic mechanisms
and the use of linear analysis tools for frequency domain measures of HRV and spectral
analysis by fast Fourier transform (FFT), and describes some results found in women.
Heart activity is largely modulated by the autonomic nervous system (ANS), which
promotes rapid adjustments in the cardiovascular system during different stimuli (i.e.,
physical exercise, mental stress and postural change) (Hainsworth, 1998). HRV is a non-
invasive measure used to analyze the influence of the autonomic nervous system on the
heart, providing information about both sympathetic and parasympathetic contributions to
consecutive heart rate (HR) oscillations. It has been proposed that a decrease in HRV is a
powerful predictor of morbidity and mortality resulting from arrhythmic complications.
HRV decreases with age (Catai et al., 2002; Melo et al., 2005) as a consequence of
parasympathetic reduction and predominance of sympathetic modulation (Lipsitz et al.,
1990; Longo & Correia, 1995; Akselrod, 1995).
The tool most commonly used in the frequency domain is spectral analysis, which consists
of decomposing the HR variation in a given period into its fundamental oscillatory
components, defining them by their frequency and amplitude. One of the mathematical
algorithms most commonly used to determine the number, frequency and amplitude of
these components is the FFT. The sum of all the components constitutes the so-called total
power spectral density. Spectral analysis involves three distinct spectral components:
1) very low-frequency (VLF) fluctuations related to the renin-angiotensin system and
thermoregulation; 2) low-frequency (LF) fluctuations related to the sympathetic and
parasympathetic nervous systems and to baroreflex activity; and 3) high-frequency (HF)
fluctuations associated with vagal activity (Longo & Correia, 1995; Task Force, 1996). The
sympathovagal balance can also be expressed by the LF/HF ratio. Based on this analysis, it
is possible to observe the predominance of one component over the other and the
relationship between them, reflecting the autonomic modulation of the heart in the control
of HR.


170
Given the importance of the autonomic nervous system to cardiovascular health, several
analytical measures, grouped into linear and non-linear methods, can be used to assess
HRV. The ECG is recorded with the subject in a steady state (when rhythms are stationary)
for a sufficiently long period to determine events occurring within the frequencies of
interest. R-R interval spectral power is calculated from this series of intervals using an
autoregressive algorithm, which yields center frequencies and absolute power of component
fluctuations (Task Force, 1996).
Sympathovagal balance (in dimensionless units) is simply the ratio of absolute LF to
absolute HF power, or the LF/HF ratio. The literature on sympathovagal balance is replete
with disclaimers that spectral power reflects fluctuations, not absolute levels of autonomic
nerve traffic (Akselrod 1995). If mathematical manipulation of R-R interval spectral power is
to inspire confidence as a robust, reliable metric, it must be grounded solidly on
physiological principles. It must stand on its own and calculations of sympathovagal
balance may obscure rather than illuminate human physiology and pathophysiology
(Eckberg 1997).
This chapter discusses the measurement and analysis of HRV, as well as results of data for
women and the relationship between aging and hormonal changes (oral contraceptives and
hormone replacement therapy), which contribute to modifications of the autonomic control
of the heart. Each item will be discussed in a separate subitem of this chapter.
2. Measurement of heart rate variability
An electrocardiogram and HR data were obtained using a one-channel heart monitor
(MINISCOPE II Instramed, Porto Alegre, RS, Brazil) and processed using a Lab. PC+ analog-
to-digital converter (Lab PC + / National Instruments, Co., Austin, TX, USA) acting as an
interface between the heart rate monitor and a microcomputer. The ECG signal was
recorded in real time after analog-to-digital conversion at a sampling rate of 500 Hz and the
R-R intervals (ms) were calculated on a beat-to-beat basis using specific software (Silva et al.,
1994). To evaluate the effect of body position on the HR response and its variability, R-R
intervals were recorded over a 15-min period under resting conditions with the subjects in
the supine and sitting positions, respectively.
HR and R-R intervals (RRI) can be obtained in real time, beat-by-beat, using the ECG and
specific software (Silva et al., 1994). First, a visual inspection of RRI (ms) distribution
obtained during 900s of collection at rest in the supine condition was carried out in order to
eliminate the fragments containing spikes, which resulted in an interval with higher stability
of ECG RRI tracing (Task Force, 1996).
3. Spectral analysis
Linear HRV can be assessed by frequency domains. For the frequency domain, a spectral
analysis was performed by FFT applied to a single window after the subtraction of a linear
trend, at the R-R intervals previously chosen. The power spectral components were obtained
at low (LF: 0.04 to 0.15 Hz) and high (HF: 0.15 to 0.4 Hz) frequencies, in absolute units (ms
2
),
and the normalized units (nu) were computed by dividing the absolute power of a given LF
or HF component (ms
2
) by the total power minus very low frequency (0.003-0.04 Hz) power

Spectral Analysis of Heart Rate Variability in Women

171
and then multiplying this ratio by 100. Since the LF band is modulated by both sympathetic
and parasympathetic activity and the HF band is correlated with vagal cardiac control, the
LF/HF ratio was calculated to determine the sympathovagal balance (Task Force, 1996).
Sympathovagal balance is the ratio between LF and respiratory-frequency powers. Based on
this analysis, it is possible to determine the predominance of one component over the other
and the relationship between them, reflecting the autonomic modulation of the heart in the
control of heart rate.
Figure 1, which is based on an autoregressive model, illustrates the HRV power spectra at
rest in the supine and sitting positions of a representative subject in different conditions.
0 0.1 0.2 0.3 0.4 0 0.1 0.2 0.3 0.4
Frequency (Hz) Frequency (Hz)
250
200
150
100
50
0
250
200
150
100
50
0
250
200
150
100
50
0
250
200
150
100
50
0
250
200
150
100
50
0
250
200
150
100
50
0
P
o
w
e
r

s
p
e
c
t
r
a
l

d
e
n
s
i
t
y

(
m
s

)
2
P
o
w
e
r

s
p
e
c
t
r
a
l

d
e
n
s
i
t
y

(
m
s

)
2
P
o
w
e
r

s
p
e
c
t
r
a
l

d
e
n
s
i
t
y

(
m
s

)
2
P
o
w
e
r

s
p
e
c
t
r
a
l

d
e
n
s
i
t
y

(
m
s

)
2
P
o
w
e
r

s
p
e
c
t
r
a
l

d
e
n
s
i
t
y

(
m
s

)
2
P
o
w
e
r

s
p
e
c
t
r
a
l

d
e
n
s
i
t
y

(
m
s

)
2
0 0.1 0.2 0.3 0.4 0 0.1 0.2 0.3 0.4
0 0.1 0.2 0.3 0.4
0 0.1 0.2 0.3 0.4
Frequency (Hz)
Frequency (Hz)
Frequency (Hz) Frequency (Hz)

Fig. 1. Power spectral density of heart rate variability of a representative subject from the
groups of young women (A and B), and postmenopausal women undergoing (C and D) and
not undergoing (E and F) estrogen therapy, obtained at rest in the supine and sitting positions,
respectively. Spectral components are shown as LF (0.04 to 0.15 Hz), HF (0.15 to 0.4 Hz) and
VLF (below 0.04 Hz). (adapted by Neves et al., 2007)


172
Figure 2 illustrates the analysis of the RRI (ms) of a volunteer at rest in the supine
position, using the power spectrum of the autoregressive model for a better view of the
spectral components. Three spectral frequency bands were obtained: 1) VLF,
corresponding to frequencies varying from 0 to 0.04 Hz; LF, corresponding to the interval
of 0.04 Hz to 0.15 Hz; and AF, corresponding to the interval of 0.15 Hz to 0.40Hz. The LF
and HF components are expressed in normalized units (UN) which correspond to the
percentage of the total power spectrum subtracted from the VLF component. These
components were also expressed as the ratio between the absolute areas of low and high
frequency (LF/HF ratio), which is indicative of the vagosympathetic equilibrium. Figure 2
illustrates the temporal series of the RRI corresponding to the 256 values of analysis
selected previously.
1080
1060
1040
1020
1000
980
960
940
920
900
880
50 100 150 200 250
50 100 150 200 250
Number of points
1080
1060
1040
1020
1000
980
960
940
920
900
880
R
-
R

I
n
t
e
r
v
a
l
s

(
m
s
)

Fig. 2. Temporal series of 256 values of R-R intervals (ms) of a volunteer in the supine
position
Because the HR presents fluctuations that are, in large part, periodic, a continuous
electrocardiographic record over short or long periods (24 h) and a subsequent graphical
representation of the normal R-R intervals over time (tachogram) produce a complex
undulatory phenomenon that can be decomposed into simpler waves through mathematical
algorithms, such as the FFT or the autoregressive model. This process, called spectral
analysis, enables the electrocardiographic signal from the temporal series (tachogram) to be
decomposed into its different frequency components, i.e., into so-called frequency bands. It
should be noted that frequency refers to the number of times a given phenomenon (e.g., a


173
sound wave, electric current or any form of cyclic wave) occurs over time. Normally, the
frequency unit employed is Hertz (Hz), which is equivalent to one cycle per second. Figure 3
shows the application of an autoregressive model to view the power spectrum of the
analysis of heart rate variability corresponding to these values of a volunteer of this study.
In long records (24 h), the total power is decomposed into four distinct bands: 1) high
frequency band (HF), oscillating at a frequency of 0.15 a 0.40 Hz, i.e., 9-24 cycles/min,
corresponding to the heart rate variations related to the respiratory cycle (respiratory sinus
arrhythmia), which are typically modulated by parasympathetic activity; 2) low frequency
or LF band (0.04 to 0.15 Hz or 2.4 to 9 cycles/min), modulated by both sympathetic and
parasympathetic activities, with a predominance of sympathetic in some specific situations,
and which reflects the oscillations of the baroreceptors system; 3) very low frequency or VLF
band (0.003 to 0.04 Hz or 0.2 to 2.4 cycles/min), depending on the thermoregulatory
mechanisms and the renin-angiotensin system, which is also regulated by sympathetic and
parasympathetic activities; and 4) ultra low frequency or ULF band (< 0.003 Hz or < 0.2
cycles/min), which corresponds to most of the total variance, but whose physiological
significance is not yet well defined. This band is influenced by the parasympathetic and
sympathetic systems and is obviously absent from short duration records. It appears to be
related with the neuroendocrine system, circadian rhythm, and other systems (Task Force,
1996).
A high frequency component equivalent to 0.25 Hz (15 cycles/min = 15 cycles/60 s = 0.25
cycles/s = 0.25 Hz), a low frequency component equivalent to 0.1 Hz (6 cycles/min) and a
very low frequency component of 0.016 Hz (1 cycle/min). The combination of these three
sine waves generates a complex wave signal that can be compared to the signal obtained
when the heart rate is expressed on a temporal graph (tachogram). Moreover, the
calculation of the area covered by each frequency band (which is proportional to the square
of the amplitude of the original signal and hence, in this case, is expressed in ms
2
) enables
one to separate the amount of variance (power) ascribed to each frequency. This allows for a
more detailed study of the individual participation of each of the divisions of the ANS
(sympathetic and parasympathetic) in different physiological and pathological situations, as
well as its relationship with the main systems that interfere with HRV (respiratory,
vasomotor, thermoregulatory, renin-angiotensin and central nervous systems). In fact, this is
the main difference between spectral analysis and time domain analysis, since the latter
generally fails to distinguish the dominant rhythms or oscillations that give the heart rate its
variability (Task Force, 1996).
Spectral components are usually measured in absolute values of power (ms
2
). However, the
values of LF and HF can also be expressed in normalized units (nu), which represent the
value of each of these components in relation to the total power (TP) minus the VLF
component. These values are calculated by means of the following formulas: HF (nu) =
HF/(TP VLF) x 100 and LF (nu) = LF/(TP VLF) x 100. This minimizes the effects of
changes in the VLF range on the other two components with faster frequencies (LF and HF).
Another frequently used measure is the LF/HF ratio, which can provide useful information
about the balance between the sympathetic and parasympathetic systems. It should also be
noted that, because absolute values in ms
2
are highly variable and distributed
asymmetrically, they usually require logarithmic transformation (Task Force, 1996).


174
150
100
50
0
P
o
w
e
r

s
p
e
c
t
r
a
l

d
e
n
s
i
t
y

(
m
s

)
2
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
Frequency (Hz)

Fig. 3. Power spectrum of the analysis of HRV obtained by applying an autoregressive
model to a dataset of 256 values of R-R intervals in the supine position from one of the
volunteers of this study, showing the VLF (light gray), LF (medium gray) and HF (dark
gray) bands
4. Heart rate variability and oral contraceptives
Third-generation combined oral contraceptives (COCs) containing desogestrel and
gestodene (GEST) were introduced to reduce adverse effects such as fluid retention, nausea,
headaches, and weight changes (Arangino et al., 1998; Read, 2010). The balance of risks and
benefits of COC use varies, depending on patterns of usage and background risk of disease
(Hannaford et al., 2010). The repercussions of COCs on cardiac autonomic modulation have
not yet been thoroughly investigated. Studies reveal that female sex hormones influence
cardiovascular autonomic function (Minson et al., 2000; Neves et al., 2007; Carter et al.,
2009). Leicht et al. (2003) reported a positive correlation between circulating estrogen levels
and HRV.
Furthermore, the cardioprotective effects of endogenous estrogen through vasodilation
and inhibition of blood vessel injuries have been reported (Mendelsohn & Karas, 1999).
Low levels of estrogen are associated with a reduction of cardiac autonomic modulation
(Moodithaya 2009). Large clinical trials have shown that the long-term use of estrogen
in combination with a progestogen may not be beneficial, and could even compromise
the efficiency of autonomic HR modulation. Minson et al. (2000) confirmed that COC use
can modify baroreflex sensitivity and sympathetic activity. However, Santos et al. (2008)
and Schueller et al. (2006) found that COC users and non-users showed similar HRV
indices.


175
Carter et al. (2009) observed no effects of OC use on the sympathetic modulation of the heart
during orthostatic stress, nor differences in that regard between the phase of intake of active
pills and that of intake of inactive pills. Women with greater physical activity, both users
and non-users of OCs, showed a predominance of parasympathetic modulation and
presented a greater complexity of pattern distribution and less regularity and predictability
of sequential patterns than sedentary groups. Wenner et al. (2006) evaluated amenorrheic
and eumenorrheic athletes who were users and non-users of OCs, and observed no
influence on cardiac autonomic function. However, other studies suggest that there is a
relationship between OC use and autonomic HR modulation, which the authors attribute to
changes in vagal peripheral modulation caused by high levels of circulating estrogen
(Minson, 2000; Leicht et al., 2003).
Santos et al. (2008) analyzed the autonomic modulation of HR based on frequency domain
(LF, HF and LF/HF) indices and found that the use of contraceptives did not affect the
results, since they detected no difference among the groups under study. This finding may
be attributed to the pharmacological properties of low estrogen/progesterone dosages, as
well as to the maintenance of the integrity of the autonomic modulation of HR, since the
values found here fall within the range of normality. The results of this study suggest that
low estrogen/progesterone dosages do not impair autonomic modulation in the age group
under study.
5. Heart rate variability and hormonal therapy
The aging process causes changes in the autonomic modulation of the cardiovascular
system, and particularly in HR. The literature reports that parasympathetic activity in the
sinus node decreases with age, leading to a reduction in HRV and a greater risk for
cardiovascular events (Lipsitz et al., 1990). Structural and functional changes in the blood
vessels, in the cardiac conduction system and in the sensitivity of baroreceptors, as well as
increased myocardial stiffness, leading to greater force of contraction and reduced
ventricular filling, contribute to reduce the functional capacity of the cardiovascular and
hemodynamic system (Walsh, 1987). In addition, with increasing age, submaximal physical
activity and decline in functional capacity lead to increased physiological stress (Perini et al.,
2002).
The incidence of cardiovascular diseases among premenopausal women is low when
compared to that of men in the same age group, but increases significantly after this period
(Gensini et al., 1996). In several countries, cardiovascular diseases are the major cause of
morbidity and mortality among postmenopausal women, representing an important public
health problem (Mosca et al., 1997). The increase in the incidence of cardiovascular events
among middle-aged women has been associated with the hypoestrogenism typical of this
period of womens lives (Greendale et al., 1999).
With regard to autonomic heart function, some studies have demonstrated the harmful
effects of hypoestrogenism on HRV. Mercuro et al. (2000) found a reduction in HRV
indices, analyzed in the time and frequency domains, after bilateral oophorectomy, i.e.,
through the interruption of estrogen production, as occurs in menopause. Liu et al. (2003)
demonstrated higher values of HRV, analyzed in the time domain, in premenopausal


176
women than in postmenopausal women and men in the same age group, illustrating the
importance of estrogens in the autonomic differences brought about by menopause.
Similar findings, also analyzed in the time domain, were reported by Brockbank et al.
(2000) for premenopausal women compared to women after more than one year of
menopause. Davy et al. (1998) reported that young women have higher HRV than
menopausal women, and that HRV in both active and sedentary women tends to decline
with advancing age. In earlier studies conducted in our laboratory (Ribeiro et al., 2001;
Neves et al., 2007), lower levels of HRV in menopausal women compared to young
women were also recorded.
To ascertain if a physical training program could promote physiological adaptations and
improved sympathovagal balance of the heart, attenuating the deleterious effects of
menopause on the cardiovascular system, Sakabe (2007) evaluated 18 sedentary women
divided into two groups: Control Group 10 postmenopausal women (50 to 60 years old)
without hormone therapy (HT); and HT Group 8 postmenopausal women (50 to 60 years
old) undergoing HT (estradiol plus levonorgestrel). Both groups were assessed at two
different times: before (assessment) and after (reassessment) a 3-month physical training
program (PTP). Protocol 1 to evaluate the autonomic modulation of the HR, the HR was
recorded under resting conditions, supine and sitting positions, for 15 minutes in each
position. The indices evaluated in Protocol 1 were: mean HR and R-R intervals (RRI),
RMSSD index of the RRI, low (LF) and high (HF) frequency bands of the spectral analysis, in
normalized units, and LF/HF ratio. It was concluded that hormone replacement therapy did
not have a significant effect on HRV.
6. Heart rate variability and menopause
Postmenopausal women have greater sympathetic and less parasympathetic activity than
premenopausal women (Brockbank et al., 2000; Earnest et al., 2010). Moreover, Mercuro et
al.s study (2000) reveals the harmful effects of hypoestrogenism on the autonomic
modulation of the HR, while other studies have demonstrated numerous evidences that
endogenous hormones (estrogen and progesterone) contribute to a cardioprotective
phenotype in women (Vitale et al., 2009).
Parasympathetic modulation shifts to a lower range with normal aging. Although
parasympathetic modulation is generally higher in women than men, aging reduces the
difference between genders, with changes in HRV beginning approximately at menopause
(Earnest et al., 2010). Boettger (2010) examined changes in cardiovascular autonomic
parameters obtained from short-term recordings over time. The data he collected indicated a
lifelong shift in autonomic balance toward sympathetic predominance, starting at the age of
30 years.
Zuttin (2009) evaluated and compared autonomic modulation of the HR at rest in healthy
young, premenopausal and postmenopausal women leading a sedentary lifestyle, to verify
cardiovascular adjustment in response to postural changes. This investigation involved 113
healthy sedentary women, who were divided into a young group (YG) with an average age
of 233.4 years (n=40), a premenopausal group (PreMG) aged 363.1 years (n= 39), and a
postmenopausal group (PostMG) with an average age of 554.5 years (n=34).


177
In the supine position, it was found that the YG presented significantly higher values of the
HF index in absolute units (ms
2
) and lower LF values (ms
2
) and ratio than the PostMG. In
addition, the YG and PostMG showed a statistical difference in all the evaluated indices
(p<0.05), while no difference was found between the PreMG and PostMG groups (p>0.05).
In a comparison of the YG vs. PreMG and YG vs. PostMG groups in the sitting position, the
YG presented significantly higher values for the ratio (p<0.05).
With regard to the effect of postural adjustment on the autonomic HR modulation, a
comparison of the indices obtained in the supine and sitting positions revealed significant
differences (p<0.05) in all the indices. On the other hand, the PreMG groups showed a
difference in the LF/HF index (p<0.05), while the PostMG group showed no significant
difference (p<0.05).
Having calculated the regression coefficients, it was found that the straight line of the
adjusted regression indicates that, as the age of the subjects increases, it is possible to
estimate the reduction of the HF index (ms
2
). The parameters indicate mainly a reduction of
the postural change in parasympathetic modulation. With aging, the adjustment capacity
diminishes, as indicated by the delta between the supine and sitting positions.
7. Conclusions
This chapter discussed the measurement and analysis of HRV, as well as results of data for
women and the relationship between aging, hormonal changes (oral contraceptives and
hormone replacement therapy) which contribute to modifications in the autonomic control
of heart rate.
8. Acknowledgments
This chapter was financed by the Brazilian research funding agencies CNPq (National
Council for Scientific and Technological Development Process no. 370448/2007-3) and
FAPESP (So Paulo Research Foundation Process no. 2006/56788-1).
9. References
Akselrod, S. (1995). Components of heart rate variability: Basic Studies. In: Heart Rate
Variability Malik, M & Camm, AJ, (Ed.), pp. 147-163, Futura Publishing Company,
New York.
Arangino, S. et al. (1998). Effect of desogestrel-containing oral contraceptives on vascular
reactivity and catecholamine levels. Contraception, Vol. 58, pp. 289 93, ISSN 1879-
0518 (Electronic).
Boettger, M. K. et al. (2010). Influence of Age on Linear and Nonlinear Measures of
Autonomic Cardiovascular Modulation. Annals Noninvasive Electrocardiology, Vol.
15, No. 2, pp. 165174., ISSN 1542-474X (Electronic).
Brockbank, C. L. et al. (2000). Heart rate and its variability change after the menopause.
Experimental Physiology, Vol. 85, No. 3, pp. 327-330, ISSN 1469-445X (Electronic).


178
Carter, J. B., Banister, E. W. & Blaber, A. P. (2003). The effect of age and gender on heart rate
variability after endurance training. Medicine and Science in Sports and Exercise, Vol.
35, No. 8, pp. 1333-1340, ISSN 1530-0315 (Electronic).
Catai, A. M. et al. (2002). Effects of aerobic exercise training on heart rate variability during
wakefulness and sleep and cardiorespiratory responses of young and middle-aged
healthy men. Brazilian Journal of Medical and Biology Research, Vol. 35, pp. 741752,
ISSN 1414-431X (Electronic).
Collier, S. R. (2008). Sex differences in the effects of aerobic and anaerobic exercise on blood
pressure and arterial stiffness. Gender Medicine, Vol.5, No. 2, pp. 115-123, ISSN 1878-
7398 (Electronic).
Davy, K. P. et al. (1998). Elevated heart rate variability in physically active young and older
adult women. Clinical Science, Vol. 94, No. 6, pp. 579-584, ISSN 1470-8736
(Electronic).
Earnest, C. et al. (2010). Autonomic function and change in insulin for exercising
postmenopausal women. Maturitas, Vol. 65, No. 3, pp. 284291, ISSN 1873-4111
(Electronic).
Eckberg, M. D. (1997). Sympathovagal Balance: a critical appraisal. Circulation, Vol. 96, No.
9, pp. 3224-32, ISSN 1524-4539 (Electronic).
Gensini, G. F. et al. (1996). Menopause and risk of cardiovascular disease. Thrombosis
Research, Vol. 84, No. 1, pp. 1-19, ISSN 1879-2472 (Electronic).
Greendale, G. A., Lee, N. P. & Arriola, E. R. (1999). The menopause. Lancet, Vol. 353, No.
9152, pp. 571-580, ISSN 1474-547X (Electronic).
Hainsworth (1998). Physiology of the cardiac autonomic system. In: Clinical guide to cardiac
autonomic tests, Malik, M., (Ed.), pp. 51-65. Kluwer Academic Publishers, Dordrecht,
Boston, London.
Hannaford, P. C. et al. (2010). Mortality among contraceptive pill users: Cohort evidence
from Royal College of General Practitioners Oral Contraception Study. British
medical journal, Vol. 340, pp. c927, ISSN 1468-5833 (Electronic).
Leicht, A.S., Hirning, D. A. & Allen, G. D. (2003). Heart rate variability and endogenous sex
hormones during the menstrual cycle in young women. Experimental Physiology,
Vol. 3, pp. 441-446, ISSN 1469-445X (Electronic).
Lipsitz, L. A. et al. (1990). Spectral characteristics of heart rate variability before and during
postural tilt. Relations to aging and risk of syncope. Circulation, Vol. 81, pp. 1803
1810, ISSN 1524-4539 (Electronic).
Liu, C. C., Kuo, T. B., Yang, C. C. (2003). Effects of estrogen on gender-related autonomic
differences in humans. American Journal of Physiology-Heart and Circulation
Physiology, Vol. 285, No. 5, pp. 2188-2193, ISSN 1522-1539 (Electronic).
Longo, D. F. & Correia, M. J. (1995). Variabilidade da frequncia cardaca. Revista Portuguesa
de Cardiologia, Vol. 14, pp. 241-262, ISSN 0870-2551.
Melo, et al. (2005). Effects of age and physical activity on the autonomic control of heart rate
in healthy men. Brazilian Journal of Medical and Biological Research, Vol. 38, n. 9, p.
1331-1338, ISSN 1414-431X (Electronic).
Mendelsohn, M. E. & Karas, R. H. (1999). The protective effects of estrogen on the
cardiovascular system. New England Journal of Medicine, Vol. 340, pp. 1801-1811,
ISSN 1533-4406 (Electronic).


179
Mercuro, G. et al. (2000). Evidence of a role of endogenous estrogen in the modulation of
autonomic nervous system. American Journal of Cardiology, Vol. 85, pp. 787-789,
Minson et al. (2000). Sympathetic activity and baroreflex sensitivity in young women taking
oral contraceptives. Circulation, Vol. 102, pp. 1473 1476, ISSN 1524-4539
(Electronic).
Moodithaya, S. S. & Avadhany, S. T. (2009). Comparison of cardiac autonomic activity
between pre and post menopausal women using heart rate variability. Indian
Journal of Physiology and Pharmacology, Vol. 53, pp. 227 234, ISSN 0019-5499 (Print).
Mosca L, et al. (1997). Cardiovascular disease in women: a statement for healthcare
professionals from the American Heart Association. Writing Group. Circulation,
Vol. 96, No. 7, pp. 2468-2482, ISSN 1524-4539 (Electronic).
Neves, V. F. C, et al. (2007). Autonomic modulation of heart rate of young and
postmenopausal women undergoing estrogen therapy. Brazilian Journal of Medical
and Biological Research, Vol. 40, pp. 491-499, ISSN 1414-431X (Electronic).
Perini, R. et al. (2002). Aerobic training and cardiovascular responses at rest and during
exercise in older men and women. Medicine and Science in Sports and Exercise,
Vol. 34, No. 4, pp. 700-708, ISSN 1530-0315 (Electronic).
Read, C. M. (2010). New regimens with combined oral contraceptive pills-moving away
from traditional 21/7 cycles. European Journal of Contraception and Reproductive
Health Care, Vol. 15, No. 2, pp. 32-41, ISSN 1473-0782 (Electronic).
Ribeiro, T. F. et al. (2001). Heart rate variability under resting conditions in postmenopausal
and young women. Brazilian Journal of Medical and Biological Research, Vol. 34, No. 7,
pp. 871-877, ISSN 1414-431X (Electronic).
Sakabe, D. W. (2007). Efeitos do treinamento fsico sobre a modulao autonmica da
frequncia cardaca e a capacidade aerbia de mulheres ps-menopausa sem o uso
de terapia hormonal, pp. 1-170, Avariable from:
http://www.teses.usp.br/teses/disponiveis/17/17145/tde-21052008-135005/pt-
br.php
Santos, M. C. S. et al. (2008). Influence of oral contraceptive use on lipid levels and
cardiorespiratory responses among healthy sedentary women. Brazilian Journal of
Physical Therapy, Vol. 12, pp. 188 94, ISSN 1413-3555.
Schueller, P. O. et al. (2006). Effects of synthetic progestagens on autonomic tone,
neurohormones and C-reactive protein levels in young healthy females of
reproductive age. International Journal of Cardiology, Vol. 111, pp. 42 48, ISSN 1874-
1754 (Electronic).
Silva, E et al. (1994). Design of a computerized system to evaluate the cardiac function
during dynamic exercise. Physics in Medicine & Biology, Vol. 33, p. 409 abstract.
Task Force of the European Society of Cardiology and the North American Society of Pacing
and Electrophysiology (1996). Heart rate variability: standards of measurements,
physiological interpretation, and clinical use. Circulation, Vol. 93, pp. 1043-1065,
Vitale, C., Mendelsohn, M. E. & Rosano, G. M. C. (2009). Gender differences in the
cardiovascular effect of sex hormones. Nature Reviews. Cardiology, Vol. 6, pp. 532
542.


180
Walsh, R. A. (1987). Cardiovascular effects of the aging process. The American Journal of
Medicine, Vol. 82, No,1B, pp. 34-40, ISSN 1555-7162 (Electronic).
Wenner, M. M. et al. (2006). Preserved autonomic function in amenorrheic athletes. Journal of
Applied Physiology, Vol. 101, pp. 590 597, ISSN 0021-8987 (Print).
Zuttin, R. Z. (2009). Influncia da idade sobre a modulao autonmica da frequncia
cardaca e a capacidade aerbia em mulheres. pp. 1-90, Piracicaba, Brazil, Available
from https://www.unimep.br/phpg/bibdig/aluno/visualiza.php?cod=528.
0
Cortical Specication of a Fast Fourier
Transform Supports a Convolution
Model of Visual Perception
Phillip Sheridan
Grifth University
Australia
1. Introduction
Currently, the full extent of the role Fourier analysis plays in biological vision is unclear.
Although we have examples of sensory organs that perform Fourier transforms, e.g. the lens
of the eye and the cochlear, to date there is no direct empirical evidence for its implementation
in cortical architecture. However, there does exist intriguing theoretical evidence that suggests
a role for the Fourier transform in a primates primary visual cortex (area V1) which emerges
from recent developments in our knowledge of contextual modulation. This paper proposes a
newFourier transformand a specication of howthis transformhas a natural implementation
in cortical architecture. The signicance of this new Fourier transform and its specication
in neural circuitry is that it provides a plausible explanation for previously unexplained
observable properties of the primate vision system.
1.0.0.1
The spatial response properties, such as orientation tuning and spatial frequency tuning,
of neurons in area V1 have been known for some time (Schiller et al., 1976). For a
while, it was generally accepted that these tuning functions of receptive elds are largely
context-independent (De Valois et al., 1979). However, later research has demonstrated
contextual inuences from the region close to the receptive eld (Sceniak et al., 2001);
(Cavanaugh et al., 2002); (Bair & Movshen, 2004). Moreover, it has been found that this near
surround region of a receptive eld can modify receptive eld responses through suppression
(Blakemore & Tobin, 1972) and by cross-orientation facilitation effects (Sillito & Jones, 1996);
(Cavanaugh et al., 2002); (Kimura & Ohzawa, 2009). It has also been demonstrated that
long-range contextual modulation is as robust a feature of neural function in area V1 as the
extensively studied receptive eld properties of this area (Lamme, 1995). Since that time, the
evidence for long-range contextual modulation continues to grow, e.g. (Zipser et al., 1996);
(Lamme et al., 1998); (Lee et al., 1998).
1.0.0.2
Concurrent with this research establishing the empirical evidence for contextual modulation
has been research aimed at developing functional models of V1 that are consistent with
the empirical evidence. In the early 1980s the concept of convolution was employed by
David Marr (Marr & Hildreth, 1980) as a model that accounted for considerable observable
7
properties of the human vision system. Since that time further theoretical and empirical
evidence has been mounting that supports such a model. In particular, it has been shown
that response properties of neurons in area V1 are modeled by convolution of the input image
with a family of Gabor functions (Sanger, 1988). Further research has demonstrated that the
upper layers of area V1 are modeled well by a bank of Gabor lters (Grigorescu et al., 2003);
(Huang et al., 2008); (Lee & Choe, 2003); (Ursine et al., 2004); (Tang et al., 2007). A related, but
alternative, approach to the Gabor response functions to model simple and complex cells of
V1 is the use of Gaussian derivatives (Huang et al., 2009). The common denominator of these
contextual modulation models is long-range convolution. However, the issue of accepting
these state of the art computational models of contextual modulation as plausible functional
models of Layer 2/3 of V1 thus becomes one of addressing the cortical convolution conundrum,
more specically: how are the large scale convolutions required by such models accounted
for in cortical architecture?
1.0.0.3
This papers goal is to address the cortical convolution conundrum. In the process, we will
propose a new fast Fourier transform, named Generalised Overarching SHIA Fast Fourier
Transform (GOSH-FFT) and argue:
GOSH-FFT has a natural implementation in the cortical architecture of visual area V1, and
Its implementation provides a plausible cortical mechanism to account for the
convolutions implied by long-range contextual modulation.
The rest of this paper is organised as follows: Section 2 provides a description of key
neurophysiological and mathematical concepts underpinning the main thrust of this paper.
Section 3 describes the Generalised Overarching SHIA Fast Fourier Transform (GOSH-FFT).
Section 4 proposes a new interpretation of the physiology of long-range intrinsic connections
and reinterprets previously introduced physiological concepts to propose a plausible cortical
implementation of GOSH-FFT. Section 5 discusses various implications of the novel material
of this paper. Section 6 summarises and concludes the paper. Section 7 is an appendix that
contains a MatLab-like pseudo-code description of GOSH-FFT and a mathematical proof of
GOSH-FFT.
2. Background
This section reviews neurophysiological and mathematical concepts pertinent to the
understanding of GOSH-FFT and its cortical implementation.
2.1 Physiological background
The primary visual cortex (V1) has a distinctive layer structure. Inputs from the lateral
geniculate nucleus (LGN) of the thalamus arrive chiey to layer 4 in the monkey (Fitzpatrick
et al., 1985). Layers 2/3 of monkey V1 contain the rst neurons in the feedforward hierarchy
that are strongly responsive to orientated stimuli (Schiller et al., 1976); (De Valois et al.,
1979); (Parker & Hawken, 1988); (Leventhal et al., 1995). Central to understanding response
properties of neurons in V1 is the notion of the receptive eld (Hubel & Wiesel, 1962). The
receptive eld of a neuron is that region of its visual eld to which it responds most strongly.
Neurons within layers 2/3 V1 have receptive elds that are oriented. That is, they prefer
stimuli that are lines of a certain orientation, or oriented texture elements (Hubel & Wiesel,
Cortical Specication of a Fast Fourier Transform Supports a Convolution Model of Visual Perception 3
1974). The spatial and temporal frequency tuning preferences of neurons in V1 can also
be measured. The neurons response properties measured via the receptive elds resemble
spatially localized lters with a preferred orientation and spatial frequency (Schiller et al.,
1976); (Foster et al., 1985); (Mikami et al., 1986); (Edwards et al., 1995) or spatio-temporal
energy (Basole et al., 2003); (Basole et al., 2006).
2.1.0.4
The orientation preference of neurons can be mapped using optical imaging techniques and
neurological studies, which show good agreement with single cell measurements (Blasdel,
1992); and groups of neurons which act as a single unit. It has been experimentally
shown that this single unit activity of large groups of single cells are composed of 10
4
(rst order approximation) interconnected cells even in one local V1 column (Siegel, 1990).
The advantages of modeling large scale neuron activity which exhibit cohort macroscopic
organisation was shown by (Sirovitch et al., 1996). No model was presented but organising
principles for analyzing and viewing data were presented. These techniques have revealed
an intricate structure to the orientation preference map in layers 2/3. A critical feature of
these structures is the orientation pinwheel (local map), in which the orientation preference of
the neuronal population changes through the entire range of 180 degrees of orientations over
the 360 degrees of polar range of the circular pinwheel. At the centre of the pinwheel is the
singularity, which is the point at which lines of iso-orientation preference meet (Obermayer &
Blasdel, 1993).
2.1.0.5
The cortex is often called the iso-cortex because of the repeated structures of which it is
comprised (Douglas & Martin, 1991). The smallest scale of structure is the minicolumn which,
in the monkey, consists of 30 adjacent pyramidal cell shafts in layers 2/3 packed within a
diameter of 23 m (Peters & Sethares, 1996). There are approximately 20 cell bodies within a
minicolumn in layers 2/3. The next largest physical scale in V1 at which repeated structures
occur is the cortical column (Lund et al., 2003). The cortical column is 200 min diameter and
is the scale at which long-range patchy connections terminate. A number of anatomical and
functional markers repeat at a larger scale of 400 m. These include the distance between
CO blobs, the approximate periodicity of the orientation preference map, and the spatial
scale of a single ocular dominance band (Lund et al., 2003). Orientation pinwheels are also
of approximately this spatial scale. Each of these functional markers has been shown to be
closely related to the systemof patchy connections, in which like response preference connects
to like, and the inter-patch distance in V1 has this same periodicity of 400 m (Bartfeld &
Grinvald, 1992); (Malach et al., 1993); (Bosking et al., 1997). The largest spatial scale is V1 itself,
which is some 4 cm wide in the monkey. There are of the order of 10,000 CO blobs in layers
2/3 of V1 (Murphy et al., 1998), and ocular dominance bands of 120 in number (Horton &
Hocking, 1998), suggesting that the multiple response property maps with periodicity of 400
m repeat around 10,000 times over layers 2/3 of V1. The input connections from the LGN
arborize at a range of scales within layer 4C of V1. These inputs are arranged in block-like
structures at the approximate scale of an ocular dominance band in layer 4C, but at a ner
scale of approximately one column in layer 4C (Fitzpatrick et al., 1985). Further ne scale
arborizations occur at approximately the scale of one minicolumn in layer 4A. At the global
scale of the cortex, inputs from the LGN are organized into a retinotopic mapping of the
visual eld (Rolls and Cowey 1970; (Tootell et al., 1988). Connectivity into the layers 2/3
183
Cortical Specification of a Fast Fourier
Transform Supports a Convolution Model of Visual Perception
of V1 occurs via a number of anatomical routes, apart from the well described feedforward
connections from layer 4 (Fitzpatrick et al., 1985). Other routes of information transfer include
extra-striate feedback (Rockland et al., 1994); (Rockland & Vanhoesen, 1994); (Angelucci et al.,
2002), long-range intrinsic bres within V1 (Blasdel et al., 1985), as well as feedback from V1
to the lateral geniculate nucleus (Marrocco et al., 1982); (Briggs & Usrey, 2007), and diffusion
of visual signal in the retina (Kruger et al., 1975); (Berry et al., 1999).
2.1.0.6
The nest scale of axonal projections within V1 are the short-range intrinsic connections that
provide connectivity between neurons up to the range approximated by an ocular dominance
column width, or 400 m. Within V1, long-range patchy connections extend for 3 mm
within the supra-granular layers (Stettler et al., 2002) and long-range connections within the
infra-granular layers extend for up to 6 mm (Rockland & Knutson, 2001). V1 also receives
feedback from at least nine extra-striate areas (Rockland & Vanhoesen, 1994). Extra-striate
feedback is considered by most researchers to be the primary source of long-range horizontal
interactions measured in V1 (Alexander & Wright, 2006). These feedback connections are fast
conducting myelinated cortico-cortical bres, and while they traverse distances of up to 10 cm
in the monkey, the transmission delays are of the same order as intrinsic short and long-range
axons within V1 (Bringuier et al., 1999); (Girard et al., 2001). These feedback connections are
often in register with the intrinsic patchy system within V1, depending on the area of origin
(Angelucci et al., 2002); (Lund et al., 2003). The middle temporal (MT) visual area will serve
here as brief illustration of the role of extra-striate feedback in V1. Receptive eld sizes in
MT are about 10 times larger than in V1 at all eccentricities (Albright & Desimone, 1987).
Small focal injections of tracer into V1 indicate that the sizes of the feedback elds from MT
to V1 are 21-fold larger than the aggregate receptive size of the V1 injection sites (Angelucci
et al., 2002). These feedback connections are an obvious substrate for the integration of global
signals into V1 (Bullier, 2001). The local-global map hypothesis (Alexander et al., 2004) of
V1 posits a non-local inuence on the structure of local maps in V1. This hypothesis states
that the global visual map in V1 is remapped to the local map scale in V1 in the form of a
map of response properties, e.g. orientation, and in the case of the monkey, spatial frequency
preference and colour selectivity. These local maps tile the surface of V1 and each receive
inputs from a large extent of the visual eld. So rather than the local map being simply a map
of primitive visual features that apply to a point in visual space, the local map is a map of primitive
visual features as they arise in the organisation of the visual eld and become relevant to a location in
visual space. As the maximum range of contextual modulation in V1 approaches the size of the
visual eld (Alexander & Wright, 2006), the local organisation of response properties can be
inuenced by the functional properties of the global visual eld.
2.2 Mathematical background
Fundamental to the Fourier transform proposed in this paper is the Spiral Honeycomb Image
Algebra (SHIA). This is a data structure that embodies important properties of the natural
visual constraints imposed by the primate eye (Sheridan et al., 2000). In particular, SHIA
has a discrete, nite and bounded domain which mimics the distribution of photo receptors
on the retinal eld. The underlying geometry of the SHIA is a hexagonal or rectangular
lattice. In the former case, each hexagon has a designated positive integer address expressed
in base seven. The numbered hexagons form clusters of super-hexagons of size 7
n
. These
self-similar super-hexagons tile the plane in a recursively modular manner. As an example, a
super-hexagon of size 7
2
= 49 and its concomitant addressing scheme is displayed in Fig. 1 (a).
(a) Hexagonal SHIA
(b) Rectangular SHIA
Fig. 1. Displays the two-level addressing scheme of SHIA: (a) Hexagonal and (b) Rectangular.
In the latter case, each rectangle has a designated positive integer address expressed in base
ve. An example of this addressing scheme is displayed in Fig. 1 (b).
185
2.2.0.7
The importance of the SHIA addressing scheme is that it facilitates primitive image
transformations of translation, rotation and scaling. One of these transformations that has
proven to be of particular relevance to the Fourier transform is one that provides rotation and
scaling. It is referred to as mapping M10 in the notation of SHIA. Fig. 2 (a) displays an image
represented in a four level SHIA, size is 7
4
= 2401. Fig. 2 (b) represents the effect of applying
M10
2
= M100 to this image.
The critical observation to make in regard to the effect of M10 is that it produces multiple
near copies at reduced resolution of the input image. This transform will play a critical role
in the proposed FFT.
2.2.0.8
The origin of what is now called a Fourier transform dates back to 1807 when Jean Baptiste
Joseph Fourier dened the notion of representing a function as a trigonometric series. The
discrete version of a Fourier transform (DFT) for a one-dimensional signal is dened as:
F(u) =
1
N
N1
x=0
f (x)e
j2ux/N
(1)
for u = 0, ...N 1, where f (x) is a real valued function, N represents the number of elements
in the signal and j
2
= 1.
The effect of this transform is to capture the spatial relationships inherent in the signal f(x)
and express these relationships as the sum of sinusoidal function (frequency components).
Similarly, the discrete version of an inverse Fourier transform (IDFT) for a one-dimensional
signal is dened as:
f (x) =
N1
u=0
F(x)e
j2ux/N
(2)
for x = 0, ...N 1, where F(u) is the Fourier transform of the real valued function f (x), N
represents the number of elements in the signal and j
2
= 1.
The effect of this inverse Fourier transform is to take a signal in frequency domain back to the
spatial domain.
2.2.0.9
Prior to the invention of the digital computer, the Fourier series was employed as a purely
analytic tool. However, since that time, the development of a class of computationally efcient
algorithms, known as fast Fourier transforms (FFT), has meant the notion has become a useful
computational tool (VanLoan, 1992). One of the most attractive computational properties of
the FFT is its ability to process signals at higher resolution with a minimal increase in cost to
complexity. Today, most of us benet from fast Fourier transforms every day without even
knowing it as these algorithms power a vast range of electronic technology such as digital
cameras and cell phones.
2.2.0.10
The relevance of a fast Fourier transform to this paper is its relationship to the notion of
convolution. The convolution of two functions f (x) and g(x) is denoted by f (x) g(x) and
(a) Four Level SHIA
(b) M10
Fig. 2. Displays (a) an image of a duck represented on a four-level SHIA; (b) the result of
applying SHIA transform M10 twice to the image displayed in (a). There are four observable
effects: 1) multiple near copies of the input image (a), 2) each copy is rotated by the same
angle, 3) each copy is scaled by the same amount, 4) applying M10 twice to the image
displayed in (b) results in the image displayed in (a).
its discrete denition is
f (x) g(x) =

ag
f (x)g(x a) (3)
187
A well known result to researchers in the eld of signal processing is the Convolution
Theorem, which relates convolution in the spatial domain to convolution in the frequency
domain. For two functions, f (x) and g(x), let F(x) and G(x) represent the Fourier transform
of f (x) and g(x) respectively. The Convolution Theorem states that,
f (x) g(x) F(x)G(x) (4)
In other words, the convolution of two functions in the spatial domain can be achieved by the
multiplication of the functions in the frequency domain.
3. Generalised Overarching SHIA fast Fourier transform (GOSH-FFT)
In this section we propose a new fast Fourier transform that, as we will see later, possesses
the potential to be implemented in cortical architecture and thereby address the cortical
convolution conundrum. Associated with SHIA, as described in Section 2.2, is a Cooley-Tucky
type fast Fourier transform, named Generalised Overarching SHIA Fast Fourier Transform
(GOSH-FFT). This novel fast Fourier transform employs the transform M10, as described
in Section 2.2, as the critical mechanism that turns a Fourier transform into a fast Fourier
transform.
3.0.0.11
Suppose an image is represented on a SHIA of size 7
n
, where n 0. Let m 0 such that
nmodulom = 0. Let k = n m and M denote M10
m
, the composite of M10 M10 ... m times.
For(i:0:k)
1. Apply M to the input;
2. Perform a discrete Fourier transform over a sequence of sub images of size 7
m
;
3. Apply the inverse of M
i
locally.
A special case of GOSH-FFT was initially described in (Sheridan, 2007), with m = 1. The
signicance of the initial work was that it demonstrated the intrinsic connection between the
Fourier transform and primitive image transformations of translation, rotation and scaling. It
also turns out that another special case of GOSH-FFT, when n = 2m, will play a critical role in
the core hypothesis of this paper. This special case, named Particular SHIA FFT (PaSH-FFT),
is illustrated in Fig. 3.
A complete statement of Algorithm 3.0.0.11 is written in MatLab-like pseudo-code and can
be found in Section 7 along with a mathematical proof that GOSH-FFT delivers a Fourier
transform.
4. Cortical implementation of contextual modulation
In Section 1, we reviewed state of the art models of contextual modulation and concluded
that these models implied the cortical convolution conundrum. We further motivate this
conundrum by observing that as a consequence of Equation 3, a convolution of the entire
visual eld requires every minicolumn in Layer 2/3 of area V1 to receive an input from every
other minicolumn of that layer. As there just are not enough connections to convolve the visual
eld in one cortical step, the convolution problem reduces to one of determining, given the
known connectivity, what is the sequence of intermediate neurons an initial input must pass
(a) Intermediate step (b) Fourier transform
(c) Intermediate step (d) Inverse Fourier transform
Fig. 3. Displays the results of applying the special case of GOSH-FFT, that is PaSH-FFT, to
image of Fig. 2 (a), with n=4 and m=2. The four sub gures display intermediate results of
PaSH-FFT: (a) on completion of rst iteration of PaSH-FFT to Fig. 2; (b) Fourier transform on
completion of second iteration; (c) on completion of rst iteration of inverse PaSH-FFT; (d)
Inverse PaSH-FFT on completion of second iteration.
through before being output as a convolved value. With the cortical convolution conundrum
thus fully formulated, in this section we will establish a specication of a sufcient sequence of
steps to address the issue. This specication will unfold in three steps. First, we will discuss
how the SHIA transform M10 manifests in cortical architecture. We will then employ this
manifestation to demonstrate how neural circuitry accommodates PaSH-FFT. Lastly, we will
show how the cortical manifestation of PaSH-FFT supports long-range convolution.
4.1 Cortical manifestation of M10
A critical component of the fast Fourier transform, PaSH-FFT, is the transform M10.
Consequently, it is an imperative of our argument that the redistribution properties of M10 be
accounted for in the neural circuitry of the visual system. To this end we now argue that the
required effects of M10 are accounted for by the long-range properties of patchy connections
189
between columns of Layer 2/3 and similarly patchy extra-striate feedback connections to area
V1.
4.1.0.12
It has been argued that the orientation pinwheel comprises a unitary organisational structure
or local map in layer 2/3 of area V1 (Hubel & Wiesel, 1974); (Bartfeld & Grinvald, 1992);
(Blasdel, 1992). When four pinwheels are reected about their common borders, a saddle
point arises at the centre of the four pinwheels. See Fig. 4.
Fig. 4. Displays a schematic diagram of the pinwheel like structures of visual area V1,
extracted from Figure 10 page 43 of Bruce et al. (2003).
4.1.0.13
In the macaque, the preferred response properties of V1 neurons can be inuenced by activity
from a wide extent of the visual eld. A review of contextual modulation in the monkey
demonstrated contextual modulation in V1 from long-ranges in the visual eld (Alexander
& Wright, 2006). The review was compiled from a number of experimental paradigms,
including visual stimulation with long lines while the neurons receptive eld is occluded
(Fiorani et al., 1992), surround only textures (Rossi et al., 2001) and colour patches placed
distally to the neurons receptive eld (Wachtler et al., 2003). It was shown that the maximum
range of contextual modulation measurable in V1 approaches a large extent of the visual
eld relative to a neurons receptive eld size or the local cortical magnication factor. Some
experimental paradigms, such as the curve tracing effect (Roelfsema &Lamme, 1998); (Khayat
et al., 2004), relative luminance (Kinoshita & Komatsu, 2001), and texture dened boundaries
(Lee et al., 1998) show excitatory contextual modulation with tuning curves that are at out
to the maximum distance tested. The functional connectivity that underlies this long-range
contextual modulation in the monkey is likely to involve cortico-cortical feedback fromhigher
visual areas working in concert with long-range intrinsic patchy connectivity. In the monkey,
the feedback connections to V1 from higher visual areas incorporate inputs from a very large
extent of the visual eld (Angelucci et al., 2002); (Lund et al., 2003).
4.1.0.14
In the analysis that follows, the combination of patchy intrinsic connections and patchy
feedback connections are therefore assumed to enable transfer of visual information at ranges
approaching the global scale of the visual eld. Moreover, we assume that the quantity and
distribution of these connections are adequate to deliver the effects of transform M10 at the
scale of the visual eld.
4.2 Cortical manifestation of PaSH-FFT
The next step in accounting for global convolution in cortical circuitry is to explore how
PaSH-FFT manifests itself in cortical architecture. The raw data, at the lowest level of
PaSH-FFT, are complex numbers that must be multiplied and added. The rst issue to
address is to justify our assumption that the operations being performed by a neuron
could be represented as complex arithmetical operations on complex numbers. Specically,
PaSH-FFT requires that a neuron can be regarded as a mechanismcapable of representing and
manipulating complex numbers in accordance with the arithmetical operations of addition
and multiplication. There are many ways in which to interpret neuronal function in terms of
complex addition and multiplication. The model presented by (MacLennan, 1999) is adequate
for the purposes of this paper, where it is shown how the representation of complex numbers
can be encoded as the rate and relative phase of axonal impulse. From this encoding, complex
multiplication is associated with the strength of a synaptic connection as the signal passes
through it and complex addition is associated with the summing of the neuronal inputs.
Thus at the lowest level of computation in our model, we assume that the operation being
performed by a neuron can be represented as complex addition and multiplication.
4.2.0.15
In area V1, each neuron makes use of information available to it in real time. There is evidence
that contextual information is projected to widespread regions in V1 in an anticipatory
manner. Since the spatial changes in the visual eld tend to be predictable from previous
visual inputs, anticipatory contextual inputs can arrive in time to be integrated in an adaptive
manner with ongoing feedforward input. In order to express the properties of widespread
contextual integration in a more formal manner, however, we will use the mathematical
convenience of assuming that each of the distinct mathematical processes to be described
occurs in a step-wise fashion. This more constrained approach allows not only each distinct
part of the process to be formulated, but also formulates the inter-relationships between
the various sub-processes. Although it is claimed that this approach is appropriate for the
purposes of this paper, it must be acknowledged that the question of how such "contextual
integration" actually occurs in the neuronal system remains open.
4.2.0.16
At the nest scale of connectivity via short-range intrinsic connections, each neuron of a
local map is treated as if it were connected to every other neuron minicolumn of that local
map. While this is not literally true, considerations of poly-synaptic interactions at this local
scale, and the real-time, anticipatory nature of visual processing means that it is a reasonable
approximation of the functional connectivity. Consequently, we can assume that each neuron
in a local map can sum the outputs of all other neurons in that local map which have been
multiplied by unique complex numbers. We call such a collection of parallel computations a
local computation. See Fig. 5, which is a schematic diagram of a local computation.
4.2.0.17
This paper now needs to discuss three types of local computations, each of which is
determined by the interpretation of the input signal and the collection of synaptic strengths
which weight the input signal. If the input signal is the spatial domain and the weights are
associated with a set of primitive roots of unity, then the resulting local computation is a
Fourier transform, denoted F. (See Equation 1 for a denition of a Fourier transform.) If the
191
N
1
N
2
N
3
N
m
x
1
x
2
x
3
x
m
m
w
i
x
i
i=1
w
1
w
3
w
2
w
m
Fig. 5. Displays a schematic diagram of a computational unit. The circles represent neurons
and the straight lines connecting the circles represent cortical connections. Each neuron
depicted at the top of the gure outputs a value x
i
. The neuron depicted at the bottom of the
gure inputs the sum of each x
i
multiplied by weight w
i
.
input signal is the frequency domain and the weights are associated with a set of inverse
primitive roots of unity, then the resulting local computation is an inverse Fourier transform,
denoted I. (See Equation 2 for a denition of an inverse Fourier transform.) If the input signal
is a frequency domain and the weights represent Fourier components, then the resulting local
computation is a convolution in the frequency domain, denoted C. (See Equations 3 and 4.)
Table 1 provides a summary of this notation.
Symbol Local Computation Input Signal Weights
F Fourier transform spatial domain primitive roots of unity
I Inverse Fourier frequency domain primitive roots of unity
C Convolution frequency domain Fourier components
Table 1. Summary of notation and denitions of local computations
4.2.0.18
At the next higher scale of connectivity each local map is assumed to have access to ongoing
activity of every other local map via long-range patchy connections and striate-extrastriate
interactions. These provide the means by which the results of local computation can be
transported to another local map as input for a further local computation. We denote such
a projection as P to represent the class of transformations (as described in Section 4.1).
4.2.0.19
All of the components of PaSH-FFT can be considered as being built of sequences of P
projections followed by a local computation. Likewise, when we employ the term local
computation in conjunction with the symbols F, I or C, it is implied that the signal within
a local map has been sent to every other neuron in that local map through a synapse whose
strength is associated with the appropriate weight of F, I or C respectively.
4.2.0.20
Although it is commonly accepted that the cortex has a massively parallel architecture,
currently there exists no comprehensive model to describe these dynamics. The absence
of such a model means that in any particular cortical process, we cannot be sure which
aspects of the process are parallel and which are intrinsically sequential. We will employ the
following notation to show how the inherently sequential steps of PaSH-FFT can be mapped
into neural circuitry. Let the symbol denote the composition of two local computations as
follows: given arbitrary local computations A and B to operate on a signal in sequence let
(s)AB = ((s)A)B.
Note that the operator is to the right of the input signal it operates on, which is enclosed in
left and right parenthesis ().
4.2.0.21
With these concepts in hand, we can now identify the sequential steps of PaSH-FFT. In this
special case, the size of the input signal is the square of the size of the local computation
and represents two iterations of GOSH-FFT, as described in Section 3. The identication of
the sequential steps also suggests the sequence of connections that the input signal must
traverse. We now illustrate this with PaSH-FFT, given an input signal s, then the application
of PaSH-FFT would be expressed as follows:
(s)PaSH FFT = (s)P F P F P (5)
Given the assumed neural parallelism, a count of the number of components on the right hand
side of the equals sign in Equation 5, reveals that a Fourier transform of the entire visual eld
can be completed by the signal traversing a sequential path connecting ve neurons. Likewise,
an inverse Fourier transform can be delivered in cortical circuitry as follows:
(s)inversePaSH FFT = (s)P I P I P (6)
4.3 Cortical manifestation of convolution
We now progress to the issue of how convolution could be implemented in cortical
architecture. To this end, we describe the various computational constraints imposed by the
computational requirements of convolution and argue that the known cortical architecture
satises these constraints.
4.3.0.22
The key to the solution of the convolution problem in the neurological domain is provided by
the Convolution Theorem, the same one employed by numerous digital signal processing
applications. This theorem was discussed in Section 2.2. The importance of the theorem
is that the convolution of two functions in the spatial domain can be achieved by the
multiplication of the functions in the frequency domain. The implications of this theorem
to the cortical convolution conundrum are signicant. In our model, the components of the
Fourier transform of the function the input signal is to be convolved with are represented
by connection weights. Then, once the input signal has been transformed to the frequency
193
domain the required convolutions can be performed by mere multiplications. In cortical
terms, each component of the signal, in the frequency domain, must traverse a connection to
one more neuron to achieve the desired multiplication. However, the resulting convolution,
in the frequency domain, must be transformed back to the spatial domain to complete the
convolution. This is achieved with an inverse Fourier transform. Accordingly, the sequence
of connections along the path that terminates in the output of a convolved value in the spatial
domain is thus given by:
(s)Convolution = (s)P F P F C P I P I P (7)
4.3.0.23
It is assumed that each component of the input signal traverses parallel paths along the
network. Thus, the net time cost to complete a convolution is equivalent to the time required
for a component of the input signal to traverse a path connecting 10 neurons. This path is
composed of ve short-range intrinsic connections and ve long-range connections.
4.4 Analysis
The plausibility of the cortical model of convolution proposed in this paper is fundamentally
predicated on the assumptions made in its formulation. Consequently, we summarise these
assumptions along with the arguments offered to justify them before we provide an analysis
of the models parameterisation:
1. The number of long-range patchy connections is adequate to achieve a redistribution of
the global signal via transform M10. This was argued in Section 4.1 and heavily relied on
a conclusion based on a review article reported in (Alexander & Wright, 2006).
2. A rst order approximation of the number of minicolumns in a local map as 10,000. This
was discussed in Section 2.1 and relied on the work reported in (Siegel, 1990).
3. The number of short-range intrinsic connections is adequate to consider each local map as
being fully connected. This was discussed in Section 4.2 and relied on the work reported
in (Siegel, 1990).
4. A rst order approximation of the number of minicolumns in the global map is 10, 000
2
=
100 million. This was discussed in Section 2.1 and was based on the work reported in
(Murphy et al., 1998) and Assumption 2.
4.4.0.24
The rst assumption is possibly the most critical as it establishes the fundamental architectural
relationship between the local and global maps and is essential to PaSH-FFT. The second
two assumptions implied that a Fourier transform of the portion of the signal represented
in a local map would be completed by each component of the signal traversing one cortical
connection and that on completion of the rst iteration of PaSH-FFT, the global signal consists
of 10,000 local discrete Fourier transforms each of which is at the scale of a local map. Then,
on completion of the second iteration of PaSH-FFT, the 10,000 local Fourier transforms would
be transformed into a global Fourier transform of size 10, 000
2
= 100 million, which by the
fourth assumption represents the size of the global signal. From this we are able to assert
that the input spatial signal would be transformed to frequency space at a cost of the signal
traversing a path connecting four neurons. With the signal in frequency space, we employed
the Convolution Theorem to assert that with each component of the global signal traversing
one additional connection, the state of the signal would represent a convolved signal in
frequency space. This assertion was predicated on the assumption that the weight of each
of these last connections represented the Fourier weight of the appropriate gaussian. The nal
step was to transform the convolved signal back from frequency space to the spatial domain.
This was achieved with the inverse PaSH-FFT, which would be completed at the additional
cost of the signal traversing a path connecting a further ve neurons. Putting these three
steps together, we arrived at a total path length of 10 connections for the global input signal
to be transformed into a representation of global convolution of the visual eld. We also note
that this analysis accounted for a single global convolution of the input signal. However,
there will be many global convolutions required, possibly up to one for every orientation
preference and spatial frequency preference represented in a local map. Although the input
spatial signal needs only to be transformed into the frequency domain once, each distinct
convolution would require a distinct set of parallel paths to transform the signal back into
the spatial domain. Consequently, the multiple convolutions would not necessarily result
in a longer path. Accordingly, we assert that the transform, PaSH-FFT, with appropriate
parameterisation, would deliver a global convolution of the visual eld. Moreover, this output
signal is generated within the required time constraints imposed by observed contextual
modulation. Given our assumptions, the lowest number of iterations required to complete
a Fourier transform is two. Consequently, 10 represents the length of the shortest path (see
equation 7) possible to deliver a global convolution via PaSH-FFT.
5. Discussion
The signal processing literature describes many different types of fast Fourier transforms
(FFT). Although any one of them represents an alternative candidate to PaSH-FFT, the
problem to address is accounting for how they might be implemented within the known
constraints of cortical architecture. All fast Fourier transforms need to rearrange components
between their intermediate steps of multiply and add. PaSH-FFT derives its rearrangements
of components with the transform M10 that, as argued, is compatible with the distribution
and quantity of long-range cortical connections. If any other FFT could be substituted for
PaSH-FFT in the model, one would need to account for the rearrangement phase of that FFT
within the known connectivity of area V1.
5.0.0.25
Another issue worthy of some discussion pertains to the Fourier transform and the absence of
empirical evidence that would irrefutably demonstrate its cortical implementation. Part of the
explanation for this lack of evidence may be provided by the role the Fourier transform plays
in the vision process as suggested by this paper. That is, PaSH-FFT was shown to be a means
to an end (convolution), not the end itself. Consequently, the question of nding neurons
through empirical experimentation that measures response properties of neurons that closely
model the prole of a Fourier transform may remain unanswered for some time to come.
5.0.0.26
Underpinning the proposed implementation of PaSH-FFT in cortical architecture is a highly
simplistic model of the parallelism inherent in the cortex. The model employed did not
take into account at least two well accepted features of this parallel architecture. First, the
system itself somehow synchronises the ow of the signal. Second, the cortex does not
195
need computation-like synchronisation or state update. Synchronisation can be provided by
considering "Small World" relationships. (Gao et al., 2001) have shown that a "Small World"
network needs only a small fraction of long-range couplings to obtain a great improvement in
both stochastic resonance and synchronisation in network connectivity of bistable oscillators.
We suggest that the known topology of the visual cortex (Zeki, 1993) if considered as a
"Small World" network can provide the foregoing benets. They would be consistent with the
long-range and short-range connectivities of V1 to retinal neurons which have the required
bistable oscillator condition provided by on-centre or off-centre neurons responses to light
and dark and including those with colour opponency properties. The long and short-range
selectivity for connections can be dynamic based on the neuron threshold levels and spatial
frequency channels (Dudkin, 1992). The system updates a neuronal state only when new
information indicates a change in the input signal.
5.0.0.27
The cortical implementation of PaSH-FFT was discussed in Section 4.2 where it was argued
that the known connectivity of area V1 was sufcient to support its cortical implementation.
It was then argued that this implementation could deliver the required convolution in a
small number of sequential steps. However, the argument did not rule out the possibility
of an alternative mechanism that would deliver the required convolution in fewer steps
than PaSH-FFT. It would appear that without a sufciently developed model of the brains
parallelism, it is unlikely that a mathematical proof of a lower bound for the minimumnumber
of sequential steps could be produced. Currently, the only bound that we can be sure of is that
the required convolution could not be completed in one step. The question of determining the
minimum lower bound remains an open question.
5.0.0.28
The role of the frequency domain was at the heart of the solution to the cortical convolution
conundrum proposed in this paper. However, the possibility of performing the convolution
in the spatial domain without resorting to the frequency domain cannot be ruled out by any
argument presented in this paper. Although it is unclear how this could be accomplished
without resorting to a highly asymmetric model of the distribution of the connectivity of
long-range connections. In any case, the search for an explanation of how the dynamic
reconguration implied by the analysis of this paper is actually accomplished is likely to
provide many different conjectures along the way. One possible avenue in this endevour
might be provided by further tracer experiments such as those reported in (Angelucci et al.,
2002).
6. Summary and conclusion
This paper reviewed the evidence for long-range contextual modulation and concluded that
it implied cortical convolution at the scale of the visual eld. This resulted in the need to
address the problem of how such long-range convolution could be accounted for with known
cortical connectivity and within known time constraints. The paper proposed a solution to
the problem that emerged from a mathematical analysis of cortical connectivity to account for
the implied constraints of long-range convolution. In particular, it was argued that the known
distribution of the long-range patchy connections and extrastriate connections is adequate
to provide the means by which the global visual signal can be transformed into frequency
space where the convolution can be performed. The main thrust of the argument was that
these long-range connections facilitated the transformation of the signal into and out of
the frequency domain via a new fast Fourier transform named PaSH-FFT. A mathematical
proof of the most general form of this FFT, GOSH-FFT, was provided in the appendix along
with MatLab-like pseudo-code to facilitate the implementation of GOSH-FFT in computer
software.
6.0.0.29
It was shown that, to a rst order approximation, a cortical implementation of PaSH-FFT could
account for the large scale convolution implied by known models of contextual modulation.
The signicance of PaSH-FFT is that it:
represents a plausible cortical mechanism to account for long-range contextual
modulation;
suggests a theoretical explanation of how the brain might be wired to achieve large scale
Fourier analysis;
opens up the possibility of explaining other cortical processes via frequency space
computations.
It is the conclusion of this paper that the processing of the visual signal in the frequency
domain via a fast Fourier transform plays a fundamental role in primate vision.
7. Appendix
In this appendix, we present the pseudo-code for GOSH-FFT and a mathematical proof of
GOSH-FFT.
7.1 Appendix A
This section presents a formal statement of GOSH-FFT in MatLab-like pseudo-code.
Notation:
x = Complex array specifying the input signal
base = 7
, where is an integer greater than zero

n = 7
, where / , is an integer greater than zero.

denotes scalar multiplication in SHIA
01 function y = GOSH FFT(x, n, base, )
02 y x
03 f orlevel = 1 : log
base
(n
04 f m10(y, )
05 r rootsOf Unity(base
level
)
06 w m10(r, )
07 F localDFFT( f , w, level, n, base)
08 y local M10Inverse(F, level, n, base, )
09 r w
10 end
11 end
197
01 function T = m10(t, )
02 f ori = 0 : n 1
03 a convertToSHI A(i)
04 b 10
a
05 c convertFromSHI A(b)
06 T(c) t(i)
07 end
08 end
01 function F = localDFT( f, w, level, n, base )
02 subSignalSize base
level
03 numSubSignals n/subSignalSize
04 f ors = 0 : numSubSignals 1
05 tab s subSignalSize
06 f ors = 0 : subSignalSize 1
07 wLB i base
08 lB i base + tab
09 uB lB + base 1
10 f oru = lB : uB
11 F(u)
base1
x=0
f (x + lB) w(x + wLB)
12 end
13 end
14 end
15 end
01 function T = localM10Inverse( t, level, n, base )
02 subSignalSize base
level
03 numSubSignals n/subSignalSize
04 for s = 0: numSubSignals-1
05 for i =0: subSignalSize-1
06 a convertToSHI A(i)
07 b 10
a
08 c convertFromSHI A(b)
09 T(c) t(i)
10 end
11 end
12 end
7.2 Appendix B
This section presents the mathematical proof of GOSH-FFT.
Notation: The symbol % will be employed to mean modular arithmetic. Let B = 7
m
, where m
is a positive integer. N = B
p
, where p is a positive integer.
Let M denote the compound transform M10
m
from SHIA.
Let 1, e, e
2
, e
3
, ..., e
N1
represent N roots of unity, where multiplication of two arbitrary roots
of unity denoted e
i
and e
j
is dened as: e
i
e
j
= e
(i+j)%N
.
e(u) = e
(u%B)+(u/B)n
= 1, e
1
, ..., e
B1
, ..., e
(B1)n
, ..., e
(B1)n+(B1)
, for p > 1 and =
1, e
1
, ..., e
B1
, for p = 1.
F(u) = F
u%B
u/B
= F
0
0
, ..., F
B1
0
, ..., F
0
n1
, ...F
B1
n1
, denote a sequance of N Fourier components.
f (x) = f
x%B
x/B
= f
0
0
, ... f
B1
0
, ..., f
0
n1
, ..., f
B1
n1
, denote a sequence of N points in the input signal.
The proof is by induction on p. When p=1, GOSH-FFT is simply a DFT. Assume the
GOSH-FFT computes a Fourier transform for all levels less than p.
M
1
( f
x%B
x/B
) = f
0
0
, . . . , f
0
n1
, . . . , f
B1
0
, . . . , f
B1
n1
.
We now have B sub-signals, each of which is composed of n points. Then, by the induction
hypothesis, we can apply GOSH-FFT to obtain B individual transforms of the B sub-signals;
which yields:
Let
f =

F
0
0
, . . . ,

F
0
n1
, . . . ,

F
B1
0
, . . . ,

F
B1
n1
.
M(

f ) =

F
0
0
, . . . ,

F
B1
0
, . . . ,

F
0
n1
, . . . ,

F
B1
n1
M(e(u)) = e((u/B) + (u%B)n)
= 1, e
n
, . . . , e
(B1)n
, e
1
, e
n+1
, . . . , e
(B1)n+1
, . . . , e
n1
, e
2n1
, . . . , e
Bn1
Perform a local DFT on each of the n groups of B points. The general term is:
B1
q=0
F
q
u/B
e
q((u/B)+((u%B)n))%N
(8)
=
B1
q=0
n1
r=0
f
q
r
e
Br(u/B)%N
e
q(u/B)+((u%B)n))%N
(9)
=
B1
q=0
n1
r=0
f
q
r
e
Bn(u/B)%N
e
(u%B)rBn%N
e
q((u/B)+((u%B))n))%N
(10)
=
B1
q=0
n1
r=0
f
q
r
e
Br((u/B)+((u%B)n))%N
e
q((u/B)+((u%B)n))%N
(11)
=
B1
q=0
n1
r=0
f
q
r
e
(Br+q)((u/B)+((u%B)n))%N
(12)
=
N1
s=0
f
s%B
s/B
e
s((u/B+((u%B)n))%N
(13)
= F((u/B) + (u%B)n)) (14)
= F(u)
199
Although this proof is for signals of size 7
n
represented on a hexagonal lattice, it applies to the
case of a signal of size 5
n
on a rectangular lattice with a few modications, i.e. all appearances
of the number 7 needs to be replaced with the number 5.
8. Acknowlegements
The author thanks Valerie Christian McDougall for her work on improving the presentation
of this manuscript.
The source code that implements GOSH-FFT is available in a software package named
VisioLab from the following website: http://www.ict.grifth.edu.au/sheridan/
9. References
Albright, T. D. &Desimone, R. (1987). Local precision of visuotopic organization in the middle
temporal area (Mt) of the macaque. Experimental Brain Research, Vol. 65, No. 3, 1987,
582-592
Alexander, D. M. & Bourke, P. D. (2004). Intrinsic connections in tree shrew V1 imply a global
to local mapping. Vision Research, Vol. 44, No. 9, 2004, 857-876
Alexander, D. M. & Wright, J. J. (2006). The maximum range and timing of excitatory
contextual modulation in monkey primary visual cortex. Visual Neuroscience, Vol. 23,
No. 5, 2006, 721-728
Angelucci, A. & Levitt, J. B. (2002). Circuits for local and global signal integration in primary
visual cortex. Journal of Neuroscience, Vol. 22, No. 19, 2002, 8633-8646
Bair, W. & Movshon, J. A. (2004) Adaptive temporal integration of motion in
direction-selective neurons in macaque visual cortex. J Neurosci., Vol. 24, No. 33, 2004,
7305-7323
Bartfeld, E. & A. Grinvald (1992). Relationships between Orientation-Preference Pinwheels,
Cytochrome-Oxidase Blobs, and Ocular-Dominance Columns in Primate Striate
Cortex. Proceedings of the National Academy of Sciences of the United States of America,
vol. 89, No. 24, pp. 11905-11909, 1992
Basole, A. & Kreft-Kerekes, V. (2006). Cortical cartography revisited: a frequency perspective
on the functional architecture of visual cortex. Visual Perception, Part 1, Fundamentals
of Vision: Low and Mid-Level Processes in Perception Vol. 154, 2006, 121-134
Basole, A. & E. White, L. (2003). Mapping multiple features in the population response of
visual cortex. Nature Vol. 23, No. 6943, 2003, 986-990
Berry, M. J. & Brivanlou, I. H. (1999). Anticipation of moving stimuli by the retina. Nature, Vol.
398, No. 6725, 1999, 334-338
Blakemore, C. & Tobin, E. A. (1972). Lateral inhibition between orientation detectors in cats
visual-cortex. Exp Brain Res., Vol. 15, No. 4), 1972, 439-441
Blasdel, G. G. & Lund, J. S. (1985). Intrinsic Connections of Macaque Striate Cortex - Axonal
Projections of Cells Outside Lamina 4C. Journal of Neuroscience, Vol. 5, No. 12, 1985,
3350-3369
Blasdel, G. G. (1992). Orientation selectivity, preference, and continuity in monkey striate
cortex. Journal of Neuroscience, Vol. 12, No. 8, 1992, 3139-61
Bosking, W. H. & Zhang, Y. (1997). Orientation selectivity and the arrangement of horizontal
connections in tree shrew striate cortex. Journal of Neuroscience, Vol. 17, No. 6, 1997,
2112-2127
Briggs, F. & Usrey, W. M. (2007). A fast, reciprocal pathway between the lateral geniculate
nucleus and visual cortex in the macaque monkey. Journal of Neuroscience, Vol. 27,
No. 20, 2007, 5431-5436s
Bringuier, V. & Chavane, F. (1999). Horizontal propagation of visual activity in the synaptic
integration eld of area 17 neurons. Science, Vol. 283, No. 5402, 1999, 695-699
Bruce, V.; Green, P.R.; Georgeson, M. A. (2003). Visual perception physiology,psychology, and
ecology fourth edition, Psychology Press Taylor and Francis Group
Bullier, J. (2001). Feedback connections and conscious vision. Trends in Cognitive Sciences, Vol.
5, No. 9, 2001, 369-370
Cavanaugh, J. R.; Bair, W.; Movshon, J. (2002) A. Selectivity and spatial distribution of signals
fromthe receptive eld surround in macaque V1 neurons. J Neurophysiol., Vol. 88, No.
5, 2002, 2547-2556
De Valois, K. K. & De Valois, R. L. (1979) Responses of Striate Cortex Cells to Grating and
Checkerboard Patterns. Journal of Physiology-London, Vol. 291, Jun 1979, 483-505
Douglas, R. J. & Martin, K. A. C. (1991). A Functional Microcircuit for Cat Visual-Cortex.
Journal of Physiology-London, Vol. 440, 1991, 735-769
Dudkin, K. (1992). Cooperative neural networks underlying image description in visual
cortex. NeuROInformatics and Neurocomputers, Vol. 1, 1992, 387 - 398
Edwards, D. P. & Purpura, K. P. (1995). Contrast Sensitivity and Spatial-Frequency Response
of Primate Cortical-Neurons in and around the Cytochrome-Oxidase Blobs. Vision
Research, Vol. 35, No. 11, 1995, 1501-1523
Fiorani, M. & Rosa, M. G. (1992). Dynamic surrounds of receptive elds in primate striate
cortex: a physiological basis for perceptual completion? Proceedings of the National
Academy of Sciences U S A, Vol. 89, No. 18, 1992, 8547-51
Fitzpatrick, D. & Lund, J. S. (1985). Intrinsic Connections of Macaque Striate Cortex - Afferent
and Efferent Connections of Lamina 4C. Journal of Neuroscience, Vol. 5, No. 12, 1985,
3329-3349
Foster, K. H. & Gaska, J. P. (1985). Spatial and temporal frequency selectivity of neurones in
visual cortical areas V1 and V2 of the macaque monkey. Journal of Physiology, Vol. 365,
1985, 331-63
Gao, Z. & Hu, B. (2001). Stochastic resonance of Small World Networks. Physicas Review, Vol.
65, No. 16209, 1-4
Girard, P. & Hupe, J. M. (2001). Feedforward and feedback connections between areas V1 and
V2 of the monkey have similar rapid conduction velocities. Journal of Neurophysiology,
Vol. 85, No. 3, 2001, 1328-1331
Grigorescu, C.; Petkov N.; Westenberg, M. A. (2003) Contour detection based on nonclassical
receptive eld inhibition. IEEE Trans. Image Process., Vol. 12, No. 7, 2003, 729-739
Horton, J. C. & Hocking, D. R. (1998). Effect of early monocular enucleation upon ocular
dominance columns and cytochrome oxidase activity in monkey and human visual
cortex. Visual Neuroscience, Vol. 15, No. 2, 1998, 289-303
Huang, W.; Jia, J.; Jia, J. ((2008) Modeling contextual modulation in the primary visual cortex.
Neural Networks, Vol. 21, No. 8, 2008, 1182-1196
Huang, W.; Jiao, L.; Jia, J.; Yu, H. (2009) A neural contextual model for detecting perceptually
salient contours. Pattern Recognition Letters, Vol. 30, No. 11, 2009, 985-993
Hubel, D. H. & Wiesel, T. N. (1962). Receptive Fields, Binocular Interaction and Functional
Architecture in Cats Visual Cortex. Journal of Physiology-London, Vol. 160, No. 1, 1962,
106
201
Hubel, D. H. & Wiesel, T. N. (1974). Sequence Regularity and Geometry of Orientation
Columns in Monkey Striate Cortex. Journal of Comparative Neurology, Vol. 158, No.
3, 1974, 267-294
Khayat, P. S. & Spekreijse, H. (2004). Correlates of transsaccadic integration in the primary
visual cortex of the monkey. Proceedings of the National Academy of Sciences of the United
States of America, Vol. 101, No. 34, 2004, 12712-12717
Kimura, R. & Ohzawa, I. (2009) Time course of cross-orientation suppression in the early
visual cortex. J Neurophysiol., Vol. 101, No. 3, 2009, 1463-1479
Kinoshita, M. & Komatsu, H. (2001). Neural representation of the luminance and brightness
of a uniformsurface in the macaque primary visual cortex. Journal of Neurophysiology,
Vol. 86, No. 5, 2001, 2559-2570
Kruger, J. & Fischer, B. (1975). Shift-Effect in Retinal Ganglion-Cells of Rhesus-Monkey.
Experimental Brain Research, Vol. 23, No. 4, 1975, 443-446
Lamme, v. A. F. (1995). The neurophysiology of gure-ground segmentation in primary visual
cortex. J. Neuroscience, Vol. 15, 1995, 1605-1615
Lamme, V. A. F.; Zipser, K.; Spekreijse, H. (1998). Figure-ground activity in primary visual
cortex is suppressed by anesthesia. Proc Natl Acad Sci USA., Vol. 95, No. 6, 1998,
3263-3268
Lee, T. S.; Mumford, D.; Romero, R.; Lamme, V. A. F. (1998). The role of the primary visual
cortex in higher level vision. Vision Res., Vol. 38, No. 15, 1998, 2429-2454
Lee, H. C. & Choe, Y. (2003). Detecting salient contours using orientation energy distribution.
Proc.Internat. Joint. Conf. on Neural Networks, Vol. 1, pp. 206-211
Leventhal, A. G. & Thompson, K. G. (1995). Concomitant Sensitivity to Orientation, Direction,
and Color of Cells in Layer-2, Layer-3, and Layer-4 of Monkey Striate Cortex. Journal
of Neuroscience, Vol. 15, No. 3, 1995, 1808-1818
Lund, J. S. & Angelucci, A. (2003). Anatomical substrates for functional columns in macaque
monkey primary visual cortex. Cerebral Cortex, Vol. 13, No. 1, 2003, 15-24
MacLennan, B. J. (1999). Field computation in natural and articial intelligence. Information
Sciences International Journal, Vol. 119 1999, 73-79
Malach, R. & Amir, Y. (1993). Relationship between Intrinsic Connections and Functional
Architecture Revealed by Optical Imaging and in-Vivo Targeted Biocytin Injections
in Primate Striate Cortex. Proceedings of the National Academy of Sciences of the United
States of America, Vol. 90, No. 22, 1993, 10469-10473
Marr, H. & Hildreth, E. (1980). Theory of edge detection. Proceedings of the Royal Society, Vol.
B-207, 1980, 187-217
Marrocco, R. T. & McClurkin, J. W. (1982). Modulation of Lateral Geniculate-Nucleus Cell
Responsiveness by Visual Activation of the Cortico Geniculate Pathway. Journal of
Neuroscience, Vol. 2, No. 2, 1982, 256-263
Mikami, A. & Newsome, W. T. (1986). Motion selectivity in macaque visual cortex.
II. Spatiotemporal range of directional interactions in MT and V1. Journal of
Neurophysiology, Vol. 55, No. 6, 1986, 1328-39
Murphy, K. M. & Jones, D. G. (1998). Spacing of cytochrome oxidase blobs in visual cortex of
normal and strabismic monkeys. Cerebral Cortex, Vol. 8, No. 3, 1998, 237-244
Obermayer, K. & Blasdel, B. G. (1993). Geometry of Orientation and Ocular Dominance
Columns in Monkey Striate Cortex. Journal of Neuroscience, Vol. 13, No. 10, 1993,
4114-4129
Parker, A. J. &Hawken, M. J. (1988). Two-Dimensional Spatial Structure of Receptive-Fields in
Monkey Striate Cortex. Journal of the Optical Society of America a-Optics Image Science
and Vision, Vol. 5, No. 4, 1988, 598-605
Peters, A. & Sethares, C. (1996). Myelinated axons and the pyramidal cell modules in monkey
primary visual cortex. Journal of Comparative Neurology, Vol. 365, No. 2, 1996, 232-255
Rockland, K. S. & Knutson, T. (2001). Axon collaterals of Meynert cells diverge over large
portions of area V1 in the macaque monkey. Journal of Comparative Neurology, Vol.
441, No. 2, 2001, 134-147
Rockland, K. S. & Lund, J. S. (1983). Intrinsic Laminar Lattice Connections in Primate
Visual-Cortex. Journal of Comparative Neurology, Vol. 216, No. 3, 1983, 303-318
Rockland, K. S. & Saleem, K. S. (1994). Divergent Feedback Connections from Areas V4 and
Teo in the Macaque. Visual Neuroscience, Vol. 11, No. 3, 1994, 579-600
Rockland, K. S. & Vanhoesen, G. W. (1994). Direct Temporal-Occipital Feedback Connections
to Striate Cortex (V1) in the Macaque Monkey. Cerebral Cortex, Vol. 4, No. 3, 1994,
300-313
Roelfsema, P. R. & Lamme, V. A. F. (1998). Object-based attention in the primary visual cortex
of the macaque monkey. Nature, Vol. 395, No. 6700 376-381
Rolls, E. T. & Cowey, A. (1970). Topography of the retina and striate cortex and its relationship
to visual acuity in rhesus monkeys and squirrel monkeys. Experimental Brain Research,
Vol. 10, No. 3, 1970, 298-310
Rossi, A. F. & Desimone, R. (2001). Contextual modulation in primary visual cortex of
macaques. Journal of Neuroscience, Vol. 21, No. 5, 2001, 1698-1709
Sanger, T. D. (1988). Stereo disparity computation using Gabor lters. Biological Cybernetics,
Vol. 59, 1988, 405-418
Sceniak, M. P.; Hawken, M. J.; Shapley, R. (2001). Visual spatial characterization of macaque
V1 neurons. J Neurophysiol., Vol. 85, No. 5, 2001, 1873-1887
Schiller, P. H. & Finlay, B. L. (1976). Quantitative Studies of Single-Cell Properties in
Monkey Striate Cortex .2. Orientation Specicity and Ocular Dominance. Journal of
Neurophysiology, Vol. 39, No. 6, 1976, 1320-1333
Schiller, P. H. & Finlay, B. L. (1976). Quantitative studies of single-cell properties in monkey
striate cortex. III. Spatial frequency. J Neurophysiol, Vol. 39, No. 6, 1976, 1334-51
Sheridan, P.; Hintz, T.; Alexander, D. (2000). Pseudo invariant transformations on a hexagonal
lattice. Image Vision Comput., Vol. 18, No. 11, 2000, 907-917
Sheridan, P. E. (2007). A Method to Perform a Fast Fourier Transform with Primitive Image
Transformations. IEEE Transactions on Image Processing, Vol. 16, No. 5, 2007, 1355-1369
Siegel, R. M. (1990). Nonlinear dynamic system-theory primary visual cortical processing.
Physica, Vol. 42, No. 1-3, 1990, 385-395
Sillito, A. M. & Jones, H. E. (1996). Context-dependent interactions and visual processing in
V1. J Physiol Paris., Vol. 90, NO. 3-4, 1996, 205-209
Sirovitch, L.; Everson, R.; Kaplan, E.; Knight, B.; OBrien, E.; Orbach, D. (1996). Modelling the
functional organisation of the visual cortex. Physica, Vol. D96, 1996, 355-366
Stettler, D. D. & Das, A. (2002). Lateral connectivity and contextual interactions in macaque
primary visual cortex. Neuron, Vol. 36, No. 4, 2002, 739-750
Tang, Q., Sang, N. & Zheng, P. (2007). Extraction of salient contours from cluttered scenes.
Pattern Reccognition, Vol. 40, No. 11, 2007, 3100-3109
Tootell, R. B. H. & Hamilton, S. L. (1988). Functional-Anatomy of Macaque Striate Cortex.
Journal of Neuroscience, Vol. 8, No. 5, 1988, 1500-1530
203
Ursine, M.; La, G. E.; Cara (2004). A model of contextual interactions and contour detection in
primary visual cortex. Neural Networks, Vol. 17, No. 5-6, 2004, 719-735
Van Loan, C. (1992). Computational Frameworks for the Fast Fourier Transform, SIAM,
Philadelphia, PA.
Wachtler, T. &. Sejnowski, T. J. (2003). Representation of color stimuli in awake macaque
primary visual cortex. Neuron, Vol. 37, No. 4, 2003, 681-691
Zeki, S. (1993). A Vision of the Brain, Blackwell Scientic Publishing, Oxford.
Zipser, K.; Lamme, V.; Schiller, P. (1996). Contextual modulation in primary visual cortex. J. of
Neuroscience, Vol. 16, No. 22, 1996, 736-738
8
Spectral Analysis of Global
Behaviour of C. Elegans Chromosomes
Afef Elloumi Oueslati
1
, Imen Messaoudi
1
,
Zied Lachiri
2
and Noureddine Ellouze
1

Unit Signal, Image et Reconnaissance de Formes, Dpartement de Gnie Electrique,
1
Ecole Nationale dIngnieurs de Tunis, BP 37, Campus Universitaire,
Le Belvdre, 1002, Tunis,
2
Dpartement de Gnie Physique et Instrumentation
Institut National des Sciences Appliques et de Technologie, BP 676,
Centre Urbain Cedex, 1080, Tunis,
Tunisie
1. Introduction
Fourier analysis is one of the most useful decomposition into frequency bands to provide a
signals variations and irregularities measure. DNA spectral analysis based on Fourier
Transform contributes in the systematic search of special DNA patterns which may
correspond to biological important markers. For example, the Fourier harmonic analysis of
the occurrence of a base A can give us the corresponding frequency with amplitude and a
phase without being able to locate it in time. However it is interesting to also detect the
moments of silence of base A i.e. the moments when this base does not exist. Such a
representation of Fourier is thus limited with signals which contain transitory elements or
evolutions in their spectral contents. For these non stationary signals, the DNA sequences, to
highlight the frequency behavior, it becomes necessary to give the frequency the possibility
of changes over time. Its the time frequency analysis aim assured by the Short Time Fourier
Transform. In fact, the punctual aspect is very important to localize particular regions in
chromosomes, to characterize the beginning of a protein coding regions or a nucleosome or
its end. By depicting the frequencies by a smoothed STFT, a 2D or 3D spectrogram
representation, specific regions appear distinctly. In this paper, we are concerned with the
periodicities 3, 6, 9 and 10.5. The periodicity 3 discussed in (Anastassiou, 2001; Berger et al,
2003; Cohanim et al, 2005; Kornberg, 1977; Segal et al, 2006; Susillo et al 2003; Trifonov &
Sussman, 1980; Trifonov, 1998; Vaidyanathan & Yoon, 2004) is related with protein coding
regions (called exons) in the gene. The periodicity 10.5 is related with nucleosomes
positions in the DNA sequence and the degree of deformability of the sequence in the DNA
helix (Hayes et al, 1990; Trifonov & Sussman, 1980; Widom, 1996; Worcel et al 1981). The
periodicity 6 and 9 are specific to C. Elegans organism.
This chapter is divided in five parts. First, we expose an introduction for relevant regions on
chromosomes. In part three, we detailed the DNAs sequence analysis approach, related to
sequence global behavior problem. It exposes the spectral analysis, which follows a certain


206
methodology that generates results to highlight the periodicities studied. This analysis is
based on organisms translated into signals by three coding techniques. The algorithm steps
of this technique are detailed to mention the generation method of spectrums and
spectrograms. The fourth part deals with the study of the frequencies evolution. It present
results for smoothed STFT, as a 1D, 2D or 3D spectrogram representation. Part five
concludes this chapter.
2. The relevant regions in chromosomes
The specific succession in the bases (A, G, C, and T) constitutes the hereditary message. Each
DNA fragment involves a specific protein synthesis process. Proteins are synthesized from a
set composed of 20 different amino acids, which are determined by three bases occurring in
subsequent order. A group of three consecutive nucleotides with desoxyribose and
phosphoric group is called a codon and a total of 64 different combinations specify 20 amino
acids and three stop codons, namely TAA, TAG, and TGA. The protein synthesis (Fig.1) is
realized in two steps: (1) the transcription within which the hereditary information is copied
into the messenger RNA and, (2) the translation in which the messenger RNA is exploited
by the ribosome to form the amino acid chain. To obtain numerical data from this succession
of symbolic bases of a DNA sequence, we use binary indicator coding techniques.
genes
DNA
sequence
Transcription
exon intron
Splicing mRNA
Traduction
Peptides

Fig. 1. The proteins synthesis steps
In a DNA sequence, electron microscopy and biochemical studies have established that the
bulk of the chromatin DNA is compacting into repeating structural units, named nucleosomes.
A model of this DNA structure in such regions is proposed by Kornberg in (Kornberg, 1974,
1977). The chromatin is a dynamic structure, oscillating between the nucleosome and open
structures depending on the environmental conditions (Kornberg, 1974, 1977; Oudet et al,
1978). And each nucleosome is formed by two molecules of each histone (protein) H2A,

Spectral Analysis of Global Behaviour of C. Elegans Chromosomes

207
H2B, H3 and H4. Each nucleosome has a diameter of 12.51 nm and contains about 200 base
pairs of DNA. This number is varying according to the chromatins origin (Hayes et al 1990;
Kornberg, 1977; Oudet et al, 1978; Worcel et al 1981). In contrast a particle named
nucleosome core is invariant in its DNA content about 146 base pairs. Interesting electron
microscopic evidence elaborated in (Oudet et al, 1978) suggests that under appropriate
conditions a nucleosome could open up into two separate half nucleosomes of diameter
9.31 nm. The finding of each type of histones in the nucleosome has suggested that a
nucleosome could be made up of two symmetrical halves (Altenburger, 1976).
released nucleosomes
histones dissociation
146 nucleotide
DNA fragment
Chromatin fiber

Fig. 2. Chromatines and nucleosomes structure
In order to study the protein coding regions signals and the nucleosome regions ones, the
DNA symbolic data must be converted to DNA signals.
3. Genomic sequence analysis based on Short Fourier Transform
In order to give frequencies more precise location in time, Gabor proposes to use a Fourier
local analyze with windows. The technique consists in segmenting signal by multiplication
by sliding window of fixed length (Mallat, 1999). Each part is analyzed independently with
a classic Fourier transform to enhance frequencies behavior. The totality of these transforms
forms the short Fourier transform and precise the frequencies location in time.
Applying coding process, the numerical signals are obtained by bases succession
description as follows:

[ ] ( ) [ ] { }
, 1,.., x n x i i N = (1)
The classic discrete Fourier transform related to numerical sequence is expressed as:

[ ] ( )
2
1
0
N
j nk
N
n
X k x n e

=
=

(2)
In order to locate the signal frequencies in time, the analysis is applied to sequences parts
generated by multiplication with a sliding analysis window.


208
For this purpose, the numerical signal x[n] is divided into frames of N length. The
expression become
[ ] [ ] [ ] , .
w
x n i x n i n = (3)
When based on the binary indicator A, the equation becomes:
[ ] [ ] [ ] , .
Aw A
x n i U n i n = (4)
With i is the windows order and the n is the adopted sliding value. The windows length
must be chosen to have an appropriate number of samples to guarantee the best frequency
resolution. On each block x
w
[n], is applied a Fourier transform to determine Xw

[k],
k[0:N-1], k represents the frequency index. The FT expression associated with each frame
is as follows:
[ ] [ ]
2
1
0
,
N
j nk
i
N
n
X k x n i e

=
=

(5)
With binary indicator A coding, the equation is:
[ ] [ ]
2
1
0
,
N
j nk
N
A A
n
X k x n i e

=
=

(6)
On the basis of this expression, many representations can be obtained. The sequence is
associated to chromosome, the first analyze consists in studying the frequency global
behavior. To enhance the frequencies, we used a mean smoothed spectrum. The principle
consists in calculating the mean of the obtained spectrum of equation.
[ ] [ ]
1
0
1
N
j i
i
X k X k
N

=
=

(7)
The chromosomes are generally constituted by more than 10 Mbp, so the obtained spectrum
needs to be smoothed. A second mean of the mean spectrums is applied. The converted
DNA sequence x[n] is divided into frames of M length with an overlap m. Each of these
frames is also divided into N frames by multiplication with a sliding analysis window w[n].
On each part, a mean smoothed spectrum is generated. Finally, the mean of the spectrum for
all the parts is calculated. The final expression of the spectrum is:

[ ] [ ]
1
0
1
M
j
j
X k X k
M

=
=

(8)
3.1 The chromosomes coding techniques
This analysis aims to study the chromosomes frequency global behaviour. For this purpose,
it is important to enhance particularly the signals generated by the protein coding regions
and the nucleosome regions. Thats why, three types of coding techniques are considered:


209
- A linear coding based on Binary indicators which is related to the base succession,
- A Structural coding with Pnuc, which is an experimental coding based on the helix
deformability
- A two-dimensional coding based on Frequency Chaos Game Representation which has
submerged from the field of physics known as 'chaotic dynamical systems'
3.1.1 Binary indicators techniques
The linear coding consists in attributing a binary value for each unit of the all indicators.
Which are included in {A,T,C,G, TT, TA, GC', AAA GGG}. The marker
associated takes the value of either 1 or 0 at location n for the first character, depending on
whether or not the corresponding character group exists from the location n.
Bases binary indicator:
[ ] [ ]
b
b B
S n U n
(9)
Where:

[ ]
1
0
b
if base b is in position n
U n
else
(10)
is the binary indicator of the base B={A,T,C,G}
Considering sequence
DNA
S and [ ]
A
U n the associated bases binary indicator
DNA
S = AATCGCGACACTCATTCGG
[ ]
A
U n = 1 1 0 0 0 0 0 1 0 1 0 0 0 1 0 0 0 0 0
Two Bases binary indicator:
[ ] [ ]
bb
bb BB
S n U n
=

(11)
where
[ ]
1
0
bb
if base bb is in position n
U n
else
(12)
is the binary indicator of the base B
B={AA,AT,AC,AG,TA,TT,TC,TG,CA,CT,CC,CG,GA,GT,GC,GG}
DNA
S and [ ]
CG
U n the associated bases binary indicator
DNA
[ ]
CG
U n = 0 0 0 1 0 1 0 0 0 0 0 0 0 0 0 0 1 0
Some dinucleotides as AA, TT, TA are enhancing the ADN flexibility around histones to
constitute nucleosomes (Fig. 3).
Codons binary indicator: the three bases association called triplet or codon have a
fundamental role in the process of amino acids fabrication. For these reasons, a coding


210
technique based on these bases association is used. We adopt binary indicators to each of
the 64 codons (Table 1)
[ ] [ ] { }
, 1
cod s
S n U i i N = = (13)
where:
[ ]
1
0
cod
if the codon cod starts at position n
U i
else
(14)
is the binary indicator of the codon cod and Ns is the sequences length. This marker takes
the value of either 1 or 0 at location n for the first character depending on whether the
corresponding character exists from the location n. Lets consider the codon binary indicator
[ ]
TCG
U n .
DNA
S and the associated codons binary indicator
DNA
[ ]
TCG
U n = 0 0 1 0 0 0 0 0 0 0 0 0 0 0 1 0

Fig. 3. DNA flexibility around histones is enhanced by dinucleotide as AA, TT, TA
BASE Codon associated
A AAA, AAT, AAC, AAG, ATA, ATT, ATC, ATG
ACA, ACT, ACC, ACG, AGA, AGT, AGC, AGG
T TAA, TAT, TAC, TAG,TTA, TTT, TTC, TTG
TCA, TCT, TCC, TCG, TGA, TGT, TGC, TGG
C CAA, CAT, CAC, CAG, CTA, CTT, CTC, CTG
CGT, CGC, CGG, CCA, CCT, CCC, CCG, CGA
G GAA, GAT, GAC, GAG, GTA, GTT, GTC, GTG
GCA, GCT, GCC, GCG, GGA, GGT, GGC, GGG
Table 1. Codon associated to each base


211
3.1.2 Pnuc: the structural coding techniques
The second coding technique is the Pnuc which is based on local bending and flexibility
properties of the double helix; it is deduced experimentally from nucleosome positioning
(Pnuc). By considering the matching of both stalks (A-T and C-G) along the helix, one bases
pair defines a plane and a direction in this plane. A description of the double helix shows
the overlapping of the plans (Fig. 4). When considering that the planes are parallel, passing
between planes needs translation and rotation of 34,3 of the orientation of the connection of
the plan.

Fig. 4. A description of the double helix shows the overlapping of the plans
Now the plans are not parallel and the axis of the double helix presents curvature. By
considering the interaction between a protein, a histone and a DNAs sequence, this
interaction is stronger when the contact area between both objects is the biggest. To increase
this surface, it is necessary to roll up as much as possible the segment of DNA around the
protein, in this way, we have two properties:
If the segment of DNA is not rolled up around the protein, it is in position of equilibrium,
the curvature is static
The stalk must be flexible to allow the additional curvature around the protein. These two
properties generate the nucleosome which generates an excessive curvature of the stalk.
Each trinucleotide is replaced by its numerical value given by the Pnuc table. The
DNA
S is
then replaced by the numerical sequence
PNUC
C .
DNA
PNUC
C = 0.7 5.3 8.3 7.5 7.5 6.0 5.4 5.2 6.5 5.8 5.4 5.4 6.7 0.7 3.0 8.3 4.7


212
Trinucleotide PNUC Trinucleotide PNUC
AAA/TTT 0.0 CAG/CTG 0.042
AAC/GTT 0.037 CCA/TGG 0.054
AAG/CTT 0.052 CCC/GGG 0.060
AAT/ATT 0.07 CCG/CGG 0.047
ACA/TGT 0.052 CGA/TCG 0.083
ACC/GGT 0.054 CGC/GCG 0.075
ACG/CGT 0.054 CTA/TAG 0.022
ACT/AGT 0.058 CTC/GAG 0.054
AGA/TCT 0.033 GAA/TTC 0.030
AGC/GCT 0.075 GAC/CTG 0.054
AGG/CCT 0.054 GCA/TGC 0.060
ATA/TAT 0.028 GCC/GGC 0.0100
ATC/GAT 0.053 GGA/TCC 0.038
ATG/CAT 0.067 GTA/TAC 0.037
CAA/TTG 0.033 TAA/TTA 0.020
CAC/GTG 0.065 TCA/TGA 0.054
Table 2. The PNuc table
The signal generated from this coding for a part of chromosome is given by Fig. 5. For
clarity purpose the signal is multiplied by 10. Fig. 6 illustrate the stft method applied on
this resulting signal. First, subfigure a shows a mean spectrum for distinct window of
length 5*10
5..
The spectrum obtained needs smoothing so for the second figure (subfigure
b) a blackman smoothing window is applied on each signal part before calculating the
mean spectrum of equation 7. In the third and last figures (subfigure c and d) the equation
8 is used and the parameters chosen are: Blackman window, M=5*10
5
, N=5*10
4
and
overlap 50% for subfigure c and N=5*10
3
with overlap 10% for subfigure d. The figure
shows that meaning and smoothing are very efficient to have the best signal (subfigure d).
In this signal, the periodicity 10 is enhanced to prove that this is a characteristic of helix
flexibility.

Fig. 5. Pnuc signal of 2000 base pairs of chromosome 1 of C. Elegans genome


213

Fig. 6. Illustration of the smoothed mean spectrum applied on Pnuc signal of 2000 base pairs
of chromosome 1 of C. Elegans genome
3.1.3 Fcgr: the two dimensionnal coding techniques
The third technique is submerged from the Chaos Game Representation (CGR) images
which can forms a global signature of bio-sequences (Almeida et al, 2001; Cenac et al, 2004;
Deshavanne et al, 1999, 2000; Joseph & Sasikumar, 2006; Oliver et al 1993; Fiser et al, 1994).
The CGR paradigm is a holistic way of DNA representation. It provides a unique scatter
pictures. In 1999, H. Joel Jeffrey uses for the first time this representation for studying the
"non-randomness" of genomic sequences (Jeffrey, 1990). The CGR is an iterative algorithm
for drawing fractal images to any desired scale. It maps nucleotide sequences in the
[0,1]x[0,1] square. The four letters A, C, G and T are placed at the corners. The binary CGR
vertices are assigned to the four nucleotides as:
( ) ( ) ( ) ( ) 0, 0 , 0, 1 , 1, 1 , 1, 0
A C G T
l l l l = = = = (15)
Deriving scatter pictures, the CGRs construction algorithm consists of three steps. First, the
four letters A, C, G and T are placed at the corners of a rectangular unit square. Second, the
first point is plotted halfway between the center of the square, and the corner corresponding


214
to the first nucleotide of the sequence. Third, the new point are marked successively half
way between the previous point and the corner corresponding to the base of each nucleotide
read from the sequence (Almeida et al, 2001; Joseph 2006). A generated CGR image can be
viewed as an image of distributed dots. Subdividing the unit square into a set of square
entries of equal size n, the number of square entries obtained is equal to 2n 2n. The number
of points counted in each sub-square represents the number of occurrence of a particular n-
lengthen pattern.
For illustration, lets consider a DNA sequence { }
1 2
, ,....,
N
S S S S = of N nucleotides, the CGR
value along this sequence is defined by equation 16. The result will be a square uniformly
and randomly filled with dots.

( )
1
1
1
2
n
n n s
X X l
+
+
= +
(16)
The first point
0
X is usually placed at the center of the square having thus the coordinates
(0.5, 0.5). Then, the next point
1 n
X
+
is repeatedly placed halfway between the previous
plotted point
n
X and the segment joining the vertex corresponding to the letter
1 n
s
+
of the
sequence. Fig. 7 illustrates the construction process of CGR trajectory for sequence
"ATCGG".

Fig. 7. An illustration of CGR trajectory for sequence ATCGG
To derive the CGR plot, the following steps are taken: First place
0
X at the squares center
and the four letters at the corners as described before (subfigure 1). From center to vertex A,
mark midpoint 1 ( address A) (subfigure 2). From 1 to T, mark midpoint 2 (address AT)
(subfigure 3). From 2 to C, mark midpoint 3 (address ATC) (subfigure 4). From 3 to G, mark
midpoint 4 (address ATCG) (subfigure 5). From 4 to G, mark midpoint 5 (address ATCGG)
(subfigure 6).


215

Fig. 8. Chaos Game Representation of the C. Eleganss gene F56F11.4
By identifying local patterns displayed in the CGR square, it is possible to identify
correspondent features of DNA sequences (Yu et al, 2008). The fractal nature of this kind of
DNA representation can be observed Fig. 8. The clustering dots in the lower corners indicate
a slightly high concentration in A and T. It is known that CGR patterns depict base
composition. In fact, we divide the CGR space with a grid of size k (i.e (2
k
2
k
) pixels) and
we count occurrence in each quadrant, the frequency of k-lengthen words occurrence can be
estimated and the frequency matrix then extracted is called FCGR (Frequency Chaos Game
Representation) (Almeida et al, 2001; Deshavanne et al, 2000; Jeffrey, 1990).
The FCGR was first investigated by Deschavanne in (Deshavanne et al, 1999) and later by
Almeida in (Almeida et al, 2001). To show the frequencies of the K-tuples, a color scheme
normalized to the distribution of frequency of occurrence of associated patterns is used
(Joseph & Sasikumar, 2006; Oliver et al, 1993; Tavassoly, 2007a; Tavassoly, 2007b; Makula,
2009; Goldman, 1993; Cnac, 2006; Tino, 1999; reference 44). A grayscale color mapping may
also be used. In Fig. 9, the dinucleotide and trinucleotide frequency matrices (k ={2,3}) are
obtained for the gene F56F11.4 of C.elegans. Thus, 2
2
x2
2
=16 cells are needed for motifs of
length two and 2
3
x2
3
=64 regions to count motifs of length 3. The darker pixels represent the
most frequently used words; when the clearest ones represent the fewer used words. CGRs
were used for displaying the behavior of sub-patterns within the same input sequence and
depicting oligo_mer composition. It forms the basis for similarity and self-similarity
algorithms in a different way from traditional alignment of nucleotides.
This FCGR cannot follow the evolution of frequencies from the beginning to the end of a
given sequence. So, we propose to generate signals from FCGR. We Generate the nth-order
FCGR for the hole sequence, and we replace the reading the first n-lengthen word in the
sequence, by the correspondent frequency of the same sub-pattern in the FCGRn matrix.


216

Fig. 9. The FCGR2 (k=2) and the FCGR3 (k=3) for the gene F56F11.4 of C. Elegans
Generating signals from FCGRs was a good way to capture such variability. For this fact, a
new 1D graphical representation of DNA sequences is introduced, which provide useful
insights into local and global characteristics of genomic sequences. This novel algorithm of
DNA coding consists of computing the k
th
-order FCGR for the whole sequence and
assigning then the value of the correspondent frequency to each k-lengthen word in the
sequence. Thus allows us to follow the frequencies evolution along a given sequence. The
the obtained plot set is called FCGR
k
-signal.
Lets We consider the given sequence
DNA
S
DNA
S = TTTAAAAGCTCGCGCTAAAA
The given sequence is divided with a k-length sliding window. A set of K-frames are
obtained which are denoted by K-mers. For example when k= {2, 3, 6}, we have 2-mers
(S
DNA
), 3-mers (S
DNA
) and 6-mers (S
DNA
).
F
K
(s) is defined to be the frequencies set of the k-substrings that appear in the sequence S.
Obviously; these frequencies derive from the appropriate FCGR
k
matrices. It follows that:
F
2
(S
DNA
)= {0.1579, 0.1579, 0.1579, 0.3684, 0.3684, 0.3684, 0.1053, 0.2105, 0.1579, 0.1053, 0.1579,
0.2105, 0.1579, 0.2105, 0.1579, 0.1579, 0.3684, 0.3684, 0.3684}
F
3
(S
DNA
)= {0.1111, 0.1111, 0.1667, 0.2778, 0.2778, 0.1111, 0.1111, 0.1667, 0.1111, 0.1111, 0.1667,
0.1111, 0.1667, 0.1667, 0.1111, 0.1667, 0.2778, 0.2778}
F
6
(S
DNA
)= {0.1333, 0.1333, 0.1333, 0.1333, 0.1333, 0.1333, 0.1333, 0.1333, 0.1333, 0.1333, 0.1333,
0.1333, 0.1333, 0.1333, 0.1333}
Fig. 10. illustrates the FCGR drawings for the case of k = {2,3and 6} and the corresponding
plot sets for the considered sequence.
Fig. 10. Also shows the slightly high concentration in AA and AAA motifs in FCR2 and
FCGR3 which are expressed by the high-rise blocks in the correspondent signals.
On the signals obtained, a spectral analysis is applied to detect the frequency global
behaviour in the spectrum for each C. Elegans chromosome.


217

(a) FCGR2 (b) FCGR2-signal

(c) FCGR3 (d) FCGR3-signal

(e) FCGR6 (f) FCGR6-signal
Fig. 10. FCG representation for k=2, 3 and 6 and the FCGR_signals associated


218
3.2 The Fourier analysis method steps
The short time analysis is the technique used in order to locate specific regions in a DNA
sequence. In this purpose, a mean values of Smoothed Discrete Fourier Transform is applied
on sliding window along the DNA sequence to follow the peaks evolution for specific
frequencies points. The Fourier analysis algorithm steps are:
The converted DNA sequence x[n] is divided into frames of M length with an overlap m.
Each of these frames is also divided into N frames by multiplication with a sliding analysis
window w[n]:
[ , ] [ ] [ ]
w
x n i x n w n i n = (17)
Where i is the window index, and n the overlap. The weighting w[n] is assumed to be non
zero in the interval [0, N-1]. The frame length value N is chosen in such a way that, on the
one hand, the parameters to be measured remain constant and, on the other hand, that there
are enough samples of x[n] within the frame to guarantee reliable frequency parameter
determination. The choice of the windowing function influences the values of the short term
parameters, the shorter the window the greater his influence (Mallat, 1999). We select N and
M frame length as power of two to apply the Fast Fourier Transform algorithm.
Each weighted block x
w
[n], of the frame is transformed in the spectral domain using
Discrete Fourier Transform (DFT), in order to extract the spectral parameters X
w
[k], where k
represents the index of the frequency ([0, N-1]). The DFT of each frame (in one of M
sequence parts) is expressed as follows:

2
1
0
[ ] [ , ]
N
j nk
i
N
w w
n
X k x n i e

=
=

(18)
Using the mean values, we calculate a DFT mean value for each frame (1: M). The
expression of mean DFT is expressed as:

1
0
1
[ ] [ ]
N
j i
w w
i
Xm k X k
N

=
=

(19)
Where i correspond to the index frame of N frames ([1...N]), k is the index of the frequency
and j correspond to the index frame of M frames ([1: M]).
We constitute the matrix
( ) , [ ]
j
w
MAT j k Xm k = (20)
With these obtained values, we can constitute the matrix to represent restricted join time
frequency information, known as 2D or 3D DNA spectrograms. This 2D or 3D
representation consists of the spectrogram amplitude for a specific index periodicity in a
specific nucleotide position in the chromosome.
4. Results
The method has been applied on C. Elegans genome. The chromosomes have been divided
on 1- millions parts. The M frames have a length of 1024 bp and an overlap m=256, the


219
N frames of each M frames have length of 256 with n =128. The fig. 11 presents some
examples for the spectrum related to each of the three coding technique used. In this figure,
we show particularly the periodicities 3 and 10 which are closely depending on coding.

Fig. 11. Examples of spectrums and spectrograms generated with a mean valued
technique based on smoothed Discrete Fourier Transform applied on sliding window
along the DNA sequence parts of C. elegans genome. Two coding methods are used:
a- linear coding technique (binary indicator) (subfigure a), b- structural coding technique
(PNUC) (subfigure b)
In order to highlight the various frequencies characteristic of an organism, the tests were
carried out with various coding over various sizes of segments and various widths. The
example presented in the Table 3 presents the percentage of contribution of the
trinucleotides in the highlighting of the various characteristic frequencies at the frequencies
1/3, 1/6.5, 1/9 and 1/10. The table shows that the organism C. Elegans is rich in
periodicities and that these periodicities are raised by more than the 3/4 of these coding
technique. We notice clearly that for periodicity 3, the rate has raised more 97 %, followed
by periodicity 10 which has 90 % and periodicity 9 with 85 %. Periodicity 6.5 is a periodicity
which is very marked for this organism 70 % of code contributes to its raising. It translates
the existence with a very high rate of 6 bases groups at the periodicity 6. The majority of
these groups represent polyA, generally associated for gene purposes.


220
chromosomes
period
Ch1 Ch2 Ch3 Ch4 Ch10
P=3 96.8% 96.8% 96.8% 96.8% 96.8%
P=6.5 64% 60.9% 81.25% 73.43% 71.9%
P=9 85.9% 89% 89.1% 82.81% 79.7%
P=10 70.3% 90.6% 90.5% 90.6% 84.3%
Table 3. The proportion of contribution of the trinucleotides in the highlighting of the
various characteristic frequencies
The Fig. 11 presents some spectrum with linear coding based on binary indicator. Each
indicator contributes on a specific periodicity enhancement. The ttt binary indicator
enhances the periodicity 10 when for the indicators tta et tgg the periodicity 3 is picked up.
The 3D spectrograms give more precision on the powers spread around these periodicities.
In fact, the peaks in these frequency locations have different power values (Fig. 12).

Fig. 12. 3-D spectrograms for binary indicators coding


221
The spectrogram 3D adds a third element to the representation 2D. In addition to the
localization of the periodicities in the segment, we visualize power associated with each
peak. We can distinguish between the peaks which we can find in all the segments for a
given periodicity: 10 and 3 and the peaks which are present in certain segments and which
were eliminated by carrying out the average
The Fig. 12 is divided on 4 subfigures. Each one add to the 2D spectrograms the power
values and locations of the periodicities Enhanced: it represents the 3-D spectrograms.
Subfigures (a) and (b) are related to chromosomes2 of C. Elegans when the subfigures (c)
and (d) concern chromosome 3.
This figure shows that for the binary indicator AA, TT, AAA and TTT, the peaks
around the frequency 1/10.5 are very pronounced. The variation of the degree view angle
demonstrates that the peaks are locally spread in the chromosome part. In the literrature, it
has been demonstrated both with the biochemical and signal processing studies, that the
periodicity 10.5 related to the nucleosomes is varying. Thats why, these figures shows in
one hand that there is peaks around this periodicity and in the other hand the peaks are
spread in specific regions in the chromosome.
The Fig. 13 represents the spectrograms recovered after PNUC coding. The analysis breaks
up the chromosome made up of 15,2Mbp into 15 parts of 1Mbp. We find the localization of
the periodicity in the ends. In reality, the periodicity peaks are missed or have of very weak
power in the sequence going of 6 Mbp with 12 Mbp, it is not localized on the centromer but
it is around ends of the helix. We find it towards the position 13 Mbp until the end. In the
parts where it exists it is not continuous, it is localized in specific times lapses.
In Fig. 14 mean valued technique based on smoothed Discrete Fourier Transform was
applied along the parts 6, 9 and 13 of the chromosome 1 of C.elegans. From the 1D, 2D and
3D plots, it is observed that coding with FCGR
2
reveal the presence of both 10.5 and 3
periodicities. The peaks are spread with different values according to parts around each of
these periodicities. Each part has each own specificity. In fact, in part 9 (subfigure a) ,
periodicities 3 and 10 just submerge from the frequency behavior with peaks of modest
values. For the part 6 these periodicities have the same behavior, the specificity is the
presence of horizontal peaks around the location 750 in this part. When the part 13 is rich in
periodicities 10 and 12 and poor in periodicity 3.
For coding with FCGR-3 (Fig.15), the very pronounced peaks correspond to the 10.5
periodicity; just in the left side other peaks appear around the frequency 0.11 which
corresponds to the 9 periodicity; in the right side a few peaks occur around the frequency
1/12. The 3 periodicity disappears in the majority of the parts and when it appears, it is
present only on a few areas with very low amplitudes. In Fig 15, we can distinguish between
frequency behavior in the three parts represented. The periodicity 10 is more pronounced
for the part 16 (subfigure b) when comparing with part 9 (subfigure a) and 10 (subfigure c).
As for the hexamers coding (FCGR
6
), we find that it enhances the frequency 1/10.5; upon
rare zones the frequency 1/12 is observed (Fig.16). We clearly notice that this coding
technique enhances the periodicity 10 and his neighbor in opposition to periodicity 3. The
three parts shows different aspect of the repartition of the periodicities. In part 9 (subfigure
a), the peaks are spread in a large frequency band around periodicity. The band is
reduced for part 16 (subfigure b) to be located in two frequencies then the power is grouped
in one frequency for part 12 (subfigure c).


222

Fig. 13. Distribution of periodicity 10 for coding pnuc along chromosome 2


223
0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
10
20
30
40
50
500 1000 1500 2000 2500 3000 3500
0.1
0.2
0.3
0.4

a- Fourier analysis of part 9 of chromosome 1
0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
10
20
30
40
50
500 1000 1500 2000 2500 3000 3500
0.1
0.2
0.3
0.4

b- Fourier analysis of part 6 of chromosome 1
0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
10
20
30
40
50
60
500 1000 1500 2000 2500 3000 3500
0.1
0.2
0.3
0.4

c- Fourier analysis of part 13 of chromosome 1
Fig. 14. Examples of spectrums and spectrograms of chromosomes parts with FCGR-2
signal coding


224
0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
5
10
15
20
25
500 1000 1500 2000 2500 3000 3500
0.1
0.2
0.3
0.4

0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
5
10
15
20
25
500 1000 1500 2000 2500 3000 3500
0.1
0.2
0.3
0.4

0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
5
10
15
20
500 1000 1500 2000 2500 3000 3500
0.1
0.2
0.3
0.4

signal coding


225
0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
0.2
0.4
0.6
0.8
1
1.2
500 1000 1500 2000 2500 3000 3500
0.1
0.2
0.3
0.4

0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
0
0.5
1
1.5
500 1000 1500 2000 2500 3000 3500
0.1
0.2
0.3
0.4

0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
0
0.5
1
1.5
2
500 1000 1500 2000 2500 3000 3500
0.1
0.2
0.3
0.4

signal coding


226
A peak around the frequency 1 / 4 nearby at position 2500 corresponds to a satellite (Fig. 16
subfigure a). This frequency derives from repetitions of certain dinucleotides in the area.
The spectrogram reveals the presence of a satellite with multiple frequencies; this is
manifested clearly in the 3D graph in the form of horizontally aligned peaks colored in red,
the higher frequencies.
5. Conclusion
In This chapter, we investigate the contribution of each coding technique: the linear, the
two-dimensional and the structural one in the enhancement of the peaks related to the C.
elegans genome periodicities. For this purpose, we use a mean values of smoothed Discrete
Fourier Transform applied on sliding window along the DNA sequence to follow the peak
evolution for specific frequency points around the frequencies. We detect periodicities
around 3, 6, 9 and 10 and found periodicities 3 and 10 related respectively to genes and the
positions of the nucleosomes. First we evaluate the frequencies spread through the
chromosomes with a 1-D spectrum. Second, we consider the 2-D and 3-D DNA
spectrograms to visually detect the specific parts of chromosomes related with protein
coding regions, nucleosomes positioning regions, and other particular regions.
The time frequency analysis made it possible to follow the periodicities evolution. We
studied the contribution of a range of binary indicators for the raising of exons peak
frequency. We also studied the localization of the areas being able to form nucleosomes.
Thanks to the spectrogram with two dimensions, we visualized the localization of the areas
corresponding to periodicity 10 in the limits and not in the center of the helix. The three-
dimensional spectrogram showed that the raised peaks do not correspond to the periodicity
10 but we see clearly in certain sequences and for some indicators two lines of peaks of
variable powers around this periodicity. This result can explain the variation between 10
and 10.7 of the periodicities associated with the nucleosomes presented in the literature. It is
also observable that these peaks are alternated around two periodicities; this result could be
associated with the phenomena of chromatin compaction.
6. References
Almeida, J.S., Carrico, J.A., Maretzek, A., Noble P.A. & Fletcher M. (2001) Analysis of
genomic sequences by Chaos GameRepresentation, Bioinformatics Vol. 17, n5, pp 429
437.
Anastassiou D. (2001), Genomic Signal processing, IEEE Signal Processing Magazine, 18 (4),
pp: 8-20.
Berger J. A., Mitra S. K.& Astola J. (2003), Power spectrum analysis for DNA sequences, Proc.
of ISSPA 2003, pp 29-32, France, 1-4 July.
Cnac, P. (2006) tude statistique de squences biologiques et convergence de martingales,
PhD thesis on Applied Mathematics, Paul Sabatier University, Toulouse III, pp
1725.
Cnac P., Fayolle G., Lasgouttes J.M., (2004) Dynamical Systems in the Analysis of Biological
Sequences, research report n 5351,pp 350.
Cohanim, A.B., Kashi Y. & Trifinov E.N. (2005), Yeast Nucleosome DNA Pattern: Deconvolution
from Genome Squences of S. cerevisiae, Journal of Biomolecular Structure & Dynamics
ISSN 0739-1102, volume 22, Issue Number 6, Adenine Press, pp: 687-693.


227
Deschavanne, P., Giron, A., Vilain, J. Dufraigne, CH., & Fertil, B. (2000), Genomic Signature
Is Preserved in Short DNA Fragment, International Symposium on Bio-Informatics
and Biomedical Engineering, IEEE, pp 161167.
Deschavanne, P., Giron, A., Vilain, J., Fagot, G. & Fertil, B. (1999) Genomic signature:
characterization and classification of species assessed by chaos game representation of
sequences, Mol Biol E, Vol 16, n10, pp 13911399.
Fiser, A., Tusnady, G.E.& Simon, I.(1994) Chaos game representation of protein structures,
J.Mol Graphics, Vol 12, pp 295,302304.
Fukushima, A., Ikemurab, T., Kinouchie, M., Oshima, T., Kudod, Y., Morig, H. & Kanaya, S.
(2002) Periodicity in prokaryotic and eukaryotic genomes identified by powerspectrum
analysis, Elsevier, Gene 300, pp 203211.
Goldman, N. (1993) Nucleotide, dinucleotide and trinucleotide frequencies explain patterns
observed in chaos game representations of DNA sequences, Nucleic Acids Research
Vol.21, n10, pp 24872491.
Godsell, D.S. & Dickerson, R.E. (1994) Bending and curvature calculations in b-dna, Nucl.
Acids Res, vol 22, pp 5497-5503.
Hayes, J.J., Tullius, T.D. & wolffe, A. P. (1990) The structure of DNA in a nucleosome,
Proceedings of the National Academy of sciences of the United States of America,
vol 87 No 19, pp 7405-7409, October.
Jeffrey, H.J., (1990) Chaos game visualization of sequences, Computers & Graphics, Elsevier,
Vol.16, n1, pp 2533.
Joseph, J. & Sasikumar, R. (2006) Chaos game representation for comparison of whole genomes,
BMC Bioinformatics, Vol 7, n1, pp 1-10..
Kornberg, R.D. (1974), Chromatin structure: a repeating unit of histones and DNA. Science 184,
pp:868-871,
Kornberg, R.D. (1977), Structure of Chromatin, Annu Rev Biochem. 46 , pp 931-954.
Makula, M., (2009) Interactive visualization of oligomer frequency in DNA Computing and
Informatics, Vol. 28, pp 10011016.
Nicorici, D., Berger, J. A. , Astola, J. & Mitra, S. K. ,(2003), Finding borders between coding and
non coding DNA regions using recursive segmentation and statistics of stop codons,
Finnish Signal Processing Symposium (FINSIG'03), Tampere, Finland, pp. 231-235.
Oliver, J. L., Bernaola-Galvan, P., Guerrero, G. & Foman-Roldan, R. (1993), Entropic profiles
of DNA sequences through chaos-game-derived images, J. Theor. Biol.,Vol 160, n4, pp
457470.
Oppenheim, A. V., Schafer ,R. W. & Buck, J. R., (1999) Discrete Time Signal Processing, 2
nd

Edition, Prentice Hall.
Oudet, P., Germond, J.E., Bellard, M., Spadafora, C. & Chambon, P. (1978) Structure of
Eucaryotic Chromosomes and Chromatin, Phylosophical Transactions of the Royal
Society of London. Series B, Biological Sciences, vol 283, No 997, pp: 241-258,
Segal, E., Fondufe-Mittendorf, Y., Chen, L., Thamstrom, A., Field, Y., Moore, I. K., Wang,
J.P.Z. & widom, J. (2006) a genomic code for nucleosome positioning, nature vol 442,
pp: 772-778, August
Sussillo, D., Kundaje, A. & Anastassiou, D., 2003 Spectrogram analysis of genomes, Eurasip
Journal of Applied Signal Processing, vol. 2003, no. 4, .


228
Tavassoly, I., Tavassoly, O., Rad, M.S.R & Dastjerdi, N.M. (2007a) Multifractal Analysis of
Chaos Game Representation Images of Mitochondrial DNA, Frontiers in the
Convergence of Bioscience and Information Technologies FBIT, IEEE, pp 224229.
Tavassoly, I., Tavassoly, O., Rad, M.S.R & Dastjerdi, N.M. (2007b) Three dimensional Chaos
Game Representation of genomic sequences, Frontiers in the Convergence of
Bioscience and Information Technologies FBIT, IEEE, pp 219223.
Tino, P. (1999) Spatial representation of symbolic sequences through Iterative Function System,
IEEE, Vol 29, n4, pp 386393.
Trifonov, E. N. & Sussman, J.L. (1980) The Pitch of chromatin DNA is Reflected in its
Nucleotide Sequence, Proceedings of the National Academy of Sciences of the
United States of America, Vol 77, No 7, part 2: Biological Sciences, pp:3816-3820.
Trifonov, E. N. (1998), 3-, 10.5-, 200- and 400-base periodicities in genome sequences, Elsevier
Physica A 249, pp :511-516,.
Vaidyanathan, P. P. & Yoon, B. J. (2004) The role of signal processing concepts in genomics and
proteomics, Journal of the Franklin Institute (Special Issue on Genomics), vol. 341, pp.
111-135.
Widom, J. (1996) Short-range Order in Two Eucaryotic Genomes: Relation to chromosome
Structure J. Mol. Biol. 259 pp 579-588
Worcel, A., Strogatz, S. & Riley, D. Structure of chromatin and the linking number of DNA,
Proceedings of the National Academy of Sciences of the United States of America,
Vol 78, No 3, part 2: Biological Sciences, pp:1461-1465, 1981
Yu, Z.G., Shi, L., Xiao, Q.J. & Anh, V., (2008) Chaos game representation of genomes and their
simulation by recurrent iterated function systems, Bioinformatics and Biomedical
Engineering ICBBE, IEEE , pp 4146.
Part 3
Fourier and Helbert Transform Applications

1. Introduction
Linear spectral transform techniques such as the discrete Fourier transform and wavelet
analysis over real and complex elds have been routinely applied in the literature (Burrus
et al. (1998); Strang & Nuygen (1996)). Furthermore, extensions of these techniques over nite
elds (Blahut & Burrus (1991); Caire et al. (1993)) have led to applications in the areas of
information theory and error control coding (Blahut (2003); Dodd (2003); Sakk (2002); Wicker
(1994)). The goal of this chapter is to reviewthe Galois Field Fourier Transform, the associated
convolution theorem and its application in the eld of error control coding. In doing so, an
interesting connection will be established relating the convolution theorem over nite elds
to error control codes designed using nite geometries (Blahut (2003); Lin & Costello (1983);
Wicker (1994)).
While a complete exposition of the eld of error control would be out of context for this
chapter, we refer the interested reader to the recent characterizations of Low-Density Parity
Check (LDPC) codes (Pusane et al. (2011); Smarandache et al. (2009); Xia & Fu (2008)). Such
formulations have led to a resurgence of interest in the design (Kou et al. (2001); O.Vontobel
et al. (2005); Tang et al. (2005); Vandendriesscher (2010)) and decoding (Kou et al. (2001); Li
et al. (2010); Liu & Pados (2005); Ngatched et al. (2009); Tang et al. (2005); Zhang et al. (2010))
of nite geometry codes. The formulation in this chapter is meant to serve as a guiding
principle relating nite geometric properties to algebraic ones. The vehicle we have chosen
to demonstrate these relationships is an example from the eld of error control. In particular,
we show how a generalized Fourier-like convolution theorem can be applied as a decoding
methodology for nite geometry codes.
We begin in Section 2 by reviewing the Galois Field Fourier Transform (GFFT) followed by
an overview of error control coding in Section 3. In addition, in Section 3.1 it is demonstrated
howthe GFFT can be applied within the context of error control coding. Section 4 then goes on
to generalize these results to linear transformations using Pascals triangle as an example. The
combinatorics of such a transformation naturally lead to the design of codes derivable from

The Fourier Convolution Theorem
over Finite Fields: Extensions of
Its Application to Error Control Coding
Eric Sakk and Schinnel Small
Department of Computer Science, Morgan State University,
Baltimore, MD,
USA
9
nite geometries. Finally, Sections 5 and 6 conclude this chapter by deriving and applying the
generalized convolution theorem.
2. The Galois Field Fourier Transform
We are particulary interested in the case of nite elds where p is a prime number and
GF(p
m
) is an element of order n. The Galois Field Fourier Transform (GFFT) and its inverse
of a vector v = {v
0
, v
1
, ..., v
n1
} over GF(p) of length n can be related via the equations:
V
j
=
n1
i=0
ij
v
i
j = 0, ..., n 1
and
v
i
= (n)
1
n1
j=0
ij
V
j
i = 0, ..., n 1.
For any vector f over GF(p) where the above equations hold true, we dene
F(v) V = {V
0
, V
1
, ..., V
n1
} (1)
as the GFFT of v and
F
1
(V) = v = {v
0
, v
1
, ..., v
n1
} (2)
as the inverse GFFT of F.
Using this formulation, given two vectors
v = {v
0
, v
1
, ..., v
n1
}
w = {w
0
, w
1
, ..., w
n1
}
(3)
over GF(p) and their associated transforms
F(v) = V = {V
0
, V
1
, ..., V
n1
}
F(w) = W = {W
0
, W
1
, ..., W
n1
},
(4)
the familiar convolution theorem can be demonstrated to hold true for the nite eld case.
Specically, computing
x
j
=
n1
k=0
v
k
w
(jk)
(5)
is equivalent to computing
x
j
= F
1
(V
j
W
j
). (6)
3. Error control coding
Given a message encoded as a vector of length k over GF(p), the goal of error control coding
(ECC) is to transform the message vector into a code vector C of length n > k in a way that
causes C to be robust to errors arising over a communication channel (such as a wireless
The Fourier Convolution Theorem over Finite Fields:Extensions of its Application to Error Control Codin 3
link, ber optic cable, etc). Rather than the message vector , it is the code vector C that is
transmitted over a channel where the receiver is only able to observe a received vector

C.
Ideally, in the absence of any noise, it should be the case that

C = C. On the other hand, if
noise is present on the channel, the method used to transform (i.e. encode) the message
into the code vector C provides a way to recover from

C. The basic strategy behind ECC is,
given a message,
a. Embed a k dimensional message vector in a larger vector space of dimension n to create
the code vector C.
b. The addition of channel noise converts C into the received vector

C.
c. If the channel noise does not cause

C to be confused with other possible encodings, the
original code vector C can be recovered using some predetermined decoding scheme.
Conceptually speaking, the

C that lies within a predened noise sphere with respect to
the original C will be decoded as the (ideally) unique C; hence, can be recovered as well.
The size of the noise sphere (which is designed as part of the code) determines how many
errors can be corrected.
The general idea behind ECC then is to nd a C that minimizes ||C

C|| ; however,
numerically determining the minimum distance solution is wrought with dimensionality
issues that can lead to computational intractability. Hence, classes of codes have been devised
that relate the message encoding method to the decoding algorithm. Such algorithms are often
iterative (Blahut (2003); Lin & Costello (1983); Wicker & Kim (2003)) and converge upon the
optimal solution by exploiting the mathematical structure designed into the code.
Two important quantities in the eld of ECC are the Hamming weight and the Hamming
distance. Consider two vectors v and w of length n over GF(p).
Denition 3.1. The Hamming weight w
H
(v) of a vector v is dened as the number of non-zero
components in v.
Denition 3.2. The Hamming distance between v and w is dened as the number of components that
differ between v and w.
For example, over GF(3), assuming n = 5, v = {0 2 1 0 2} and w = {0 2 2 1 2}, according to
the above denitions we have that w
H
(v) = 3, w
H
(w) = 4 and d
H
(v, w) = 2.
An important quantity for dening the noise sphere is referred to as d
min
which is the
minimum Hamming distance between all code vectors dened in the code class. To correct
up to t errors in any code vector, it turns out that d
min
= 2t +1. Furthermore, when the ECC
is a linear code, a major simplication arises where d
min
is simply the minimum Hamming
weight computed over all non-zero code vectors in the code class.
3.1 Application of the GFFT to Reed-Solomon codes
The GFFT and the convolution theorem have been applied in the eld of error control coding
for the construction of a class of linear codes known as Reed-Solomon codes (Blahut (2003);
Wicker (1994)). The algorithm for encoding a message vector over GF(p
m
) of length k is
233
The Fourier Convolution Theorem over Finite Fields:
Extensions of Its Application to Error Control Coding
quite straightforward. To be able to correct up to t errors, create a vector of length n by
appending with 2t consecutive zeros. The code vector C is then derived by computing the
inverse GFFT of the appended construction. One approach to proving that this construction
is capable of correcting up to t errors involves applying the GFFT convolution theorem.
Specically, given a code vector C, a locator vector must be dened such that C
j
j
= 0 for
all j = 0, , n. Letting c and denote the GFFT of C and , the convolution theorem implies
c = 0. Based upon the convolution approach, the conclusion can be reached that the
inverse GFFT construction leads to Reed-Solomon codes capable of correcting up to t errors
in the code vector (Blahut (2003); Wicker (1994)).
The key feature of the GFFT approach to constructing Reed-Solomon codes described above is
that restrictions are placed on the position and the number of zeros appended to the message
vector. To summarize:
i. Addition of zeros to the message vector of length k is performed at prescribed locations.
ii. The resulting vector is then inverse transformed in order to compute the code vector C.
iii. The error correcting properties of this code can be demonstrated by applying the
convolution theorem.
In this work, one our of goals is to demonstrate that, given other linear transformations
inducing a convolution theorem, the above steps can be generalized to other classes of
codes. As we shall see, the key is to dene the transform and the structure of how zeros
are introduced into the message vector.
4. Pascal codes
4.1 The Pascal matrix over nite elds
Let us now focus our attention on the case of GF(p) where p is prime. Our starting point will
be:
Denition 4.1. Let p be a prime number, then the ij
th
entry of a p
m
p
m
m
th
order Pascal matrix
P
p
m over GF(p) is dened as
p
ij
= (j!)((j i)!i!)
1
mod p
=
j
i
mod p
(7)
for i, j = 0, 1, ..., p
m
1 and, by convention, if i > j, then p
ij
= 0.
In other words, P
p
m is an upper triangular matrix whose non-zero entries are the elements of
Pascals triangle taken mod p. For the purposes of this work, it is useful to observe that P
p
m
also has a Kronecker product description (Sakk & Wicker (2003)):
P
p
m = P
p
P
p
m1 mod p (8)
where P
p
is a 1
st
order Pascal matrix.
Example 4.2. Consider the binary case where p = 2 and m = 3. Equation (8) gives
P
2
3 =
1 1 1 1 1 1 1 1
0 1 0 1 0 1 0 1
0 0 1 1 0 0 1 1
0 0 0 1 0 0 0 1
0 0 0 0 1 1 1 1
0 0 0 0 0 1 0 1
0 0 0 0 0 0 1 1
0 0 0 0 0 0 0 1
.
Example 4.3. Consider the ternary case where p = 3 and m = 2. Equation (8) gives
P
3
2 =
1 1 1 1 1 1 1 1 1
0 1 2 0 1 2 0 1 2
0 0 1 0 0 1 0 0 1
0 0 0 1 1 1 2 2 2
0 0 0 0 1 2 0 2 1
0 0 0 0 0 1 0 0 2
0 0 0 0 0 0 1 1 1
0 0 0 0 0 0 0 1 2
0 0 0 0 0 0 0 0 1
.
4.2 The inverse of the Pascal matrix
As Section 5 will require understanding Q
p
m P
1
p
m , we introduce the following.
Observation 4.4. Let p be prime and let Q
p
be the p p matrix dened by
q
ij
=
(1)
ji
(
j
i
) mod p if j i,
0 otherwise
for i,j=0, 1, ..., p-1. (9)
Then Q
p
= P
1
p
mod p.
This result easily follows from the integer case (Call & Velleman (1993); Heller (1963)).
Furthermore, it has been demonstrated that (Sakk (2002)):
Observation 4.5. If p is prime and P
p
is a 1
st
order Pascal matrix over GF(p), then
P
p
p
mod p = I
p
(10)
where I
p
= p p identity matrix.
Hence, it easily follows that
Corollary 4.6. If p is prime and P
p
is a Pascal matrix over GF(p), then
Q
p
= P
1
p
mod p = P
p1
p
mod p. (11)
235
Example 4.7. A Pascal matrix over GF(5) and its inverse:
P
p
=
1 1 1 1 1
0 1 2 3 4
0 0 1 3 1
0 0 0 1 4
0 0 0 0 1
, Q
p
= P
4
p
=
1 4 1 4 1
0 1 3 3 1
0 0 1 2 1
0 0 0 1 1
0 0 0 0 1
.
Based upon Equation (8), it should be clear that
Q
p
m = Q
p
Q
p
m1 mod p. (12)
Finally, based upon Equation (10), it also follows that, for the m
th
order case,
P
p
p
m mod p = I
p
m (13)
where I
p
m = p
m
p
m
identity matrix. In a manner similar to the m = 1 case, this
characterization provides a path to computing the m
th
order inverse
Q
p
m P
1
p
m mod p = P
p1
p
m mod p. (14)
4.3 Error control codes designed from Pascal matrices
In a manner similar to the GFFT approach to Reed-Solomon codes summarized in Section
3.1, it has been pointed out that P
p
m can also be used to transform message vectors with the
appropriate coordinates set equal to zero (Sakk & Wicker (2003)). More precisely, we have the
following:
Denition 4.8. Consider an m
th
order Pascal matrix over GF(p) and let r be an integer such that
0 r m(p 1). Also, consider the p-ary expansion of an index
i = i
0
p
0
+ i
1
p
1
+ + i
m1
p
m1
where 0 i
j
p 1 for 0 j m1. A codeword c for an r
th
order Pascal code of length p
m
,
denoted by P
p
(r, m), is generated by
C = P
p
m (15)
where
=
0

1

p
m
1
is a message vector of length p

m
1 such that
i
GF(p),
i
= 0 i f w
p
(i) > r
i
= 0 i f w
p
(i) r
(16)
and
w
p
(i)
m1
j=0
i
j
.
Error control codes derived fromthe m
th
order Pascal matrix over GF(2) (i.e. binary data) have
been related (Forney (1988); Massey et al. (1973)) to a class of codes known as r
th
order binary
Reed-Muller codes RM(r, m) of length 2
m
(MacWilliams & Sloane (1977); Wicker (1994)). In
addition, it has been further demonstrated (Sakk (2002)) that P
2
(r, m) codes over GF(2) are
equivalent to RM(r, m) codes with minimum distance d
min
= 2
mr
. These observations have
been extended where it has been demonstrated that P
p
(r, m) codes over GF(p) are equivalent
to generalized Reed-Muller codes (GRM) codes (Sakk (2002)).
To place this class of codes in the same context as that outlined in Section 3.1, we must
show how to introduce zeros into the message vector, apply the Pascal matrix as the linear
transformation and, based upon this transformation, introduce a convolution theorem. From
the denition above, a given code is specied by choosing p, m and a value of 0 r
m(p 1). The code vector length then becomes n = p
m
; and, for this class of codes, a given
value of r denes the length k of the message. The rest of the n k components of must be
set to zero in a systematic way that leads to the minimum distance property of the code.
Example 4.9. Consider P
2
3 in Example 4.2 (hence, n = 2
3
= 8) and a message vector =
(
0
,
1
, ...,
7
) and let s be the number of consecutive zeros in the vector for a given value of
r:
r = 0 (d
min
= 8) : s = 7 = (
0
, 0, 0, 0, 0, 0, 0, 0) (k = 1)
r = 1 (d
min
= 4) : s = 3 = (
0
,
1
,
2
, 0,
4
, 0, 0, 0) (k = 4)
r = 2 (d
min
= 2) : s = 1 = (
0
,
1
,
2
,
3
,
4
,
5
,
6
, 0) (k = 7)
r = 3 (d
min
= 1) : s = 0 = (
0
,
1
,
2
,
3
,
4
,
5
,
6
,
7
) (k = 8)
Example 4.10. Consider P
3
2 in Example 4.3 (hence, n = 3
2
= 9) and a message vector =
(
0
,
1
, ...,
8
) and let s be the number of consecutive zeros in the vector for a given value of
r:
r = 0 (d
min
= 9) : s = 8 = (
0
, 0, 0, 0, 0, 0, 0, 0, 0) (k = 1)
r = 1 (d
min
= 6) : s = 5 = (
0
,
1
, 0,
3
, 0, 0, 0, 0, 0) (k = 3)
r = 2 (d
min
= 3) : s = 2 = (
0
,
1
,
2
,
3
,
4
, 0,
6
, 0, 0) (k = 6)
r = 3 (d
min
= 2) : s = 1 = (
0
,
1
,
2
,
3
,
4
,
5
,
6
,
7
, 0) (k = 8)
r = 4 (d
min
= 1) : s = 0 = (
0
,
1
,
2
,
3
,
4
,
5
,
6
,
7
,
8
) (k = 9)
In the above examples, d
min
is shown in parentheses for each value of r; furthermore, observe
that d
min
= s +1. Recalling for a moment the GFFT approach to Reed-Solomon code design,
the minimum distance of a code where the message vector has n k consecutive zeros can
be shown to be d
min
= n k + 1 (Blahut (2003); Wicker (1994)). It is apparent that, by using
a Pascal matrix as the transform, a result similar to that of the GFFT can be ascertained. The
major difference is that, for Reed-Solomon codes, the string of zeros must occur at the end of
the message vector before applying the GFFT to create C. For P(r, m), in addition to the string
of consecutive zeros, based upon the structure of P
p
m, zeros must also be dispersed in other
positions within to form code vectors C = P
p
m.
237
5. Extensions of the Fourier convolution theorem over nite elds
The convolution operation involves relating the componentwise product of two vectors in
one domain to the convolution of their transforms (Blahut & Burrus (1991)). Many linear
transforms have well-dened convolution operations. For instance, the Hadamard transform
yields the so-called logical or dyadic convolution operation (Ahmed et al. (1973); Dodd
(2003); Robinson (1972)). In this chapter, we develop extensions of the convolution theorem
that can be used to reveal useful properties of other classes of codes. As an example, we
demonstrate how the GFFT approach can be applied to describe generalized Reed-Muller
codes (Blahut (2003)).
To begin the formulation, we consider the componentwise product
j
=
j
j
of two vectors
= (
0
...
n1
) and = (
0
...
n1
). Furthermore, we consider matrix transforms such that
C P
p
m and P
p
m or, equivalently, = CQ
p
m and = Q
p
m where (P
p
m )
1
Q
p
m.
(here, denotes the message vector and C denotes the code vector). We demonstrate a
formulation analogous to the convolution operation that describes = Q
p
m:
i
=
n1
j=0
j
p
ji
mod p i = 0, 1, ..., n 1
=
n1
j=0
(
j
j
)p
ji
mod p
=
n1
j=0
j
(
n1
k=0
k
q
kj
)p
ji
mod p
=
n1
k=0
k
(
n1
j=0
j
q
kj
p
ji
) mod p
n1
k=0
k
T
i,k
mod p i = 0, 1, ..., n 1
(17)
where n = p
m
.
Notice that if we are dealing with familiar spectral transforms such as the Fourier or the
Hadamard transform (where P denotes the forward transform and Q denotes the inverse
transform), T
i,k
takes on a simple form. This is because the product q
kj
p
ji
in
n1
j=0

j
q
kj
p
ji
reduces to a term that enables us to take the transform of as C
f (i,k)
=
1
n
(
n1
j=0

j
p
j, f (i,k)
).
For the case of the Fourier transform f (i, k) = i k and T
i,k
= C
(ik)
; as expected, one ends
up with the convolution theorem (Blahut (2003); Wicker (1994)). In the case of a Hadamard
transform, f (i, k) = i k (where denotes bit-by-bit addition of the binary expansions of i
and k) and T
i,k
= C
(ik)
. Here, the bit-by-bit addition of the binary expansions of i and k
over GF(2) would result in the dyadic convolution (Ahmed et al. (1973); Robinson (1972)).
For the codes in this presentation, the q
kj
p
ji
term in the above summation leads to a
convolution theorem that depends on the matrix P
p
m. Furthermore, this theorem can also
be applied to demonstrate how to decode C to recover the message vector . In Equation (17)
q
kj
= (1)
jk
(
j
k
) mod p and p
ji
= (
i
j
) mod p; therefore, the product q
kj
p
ji
will not lead to an
expression that readily reduces the inner summation to a single term. To see why, lets write
out T
i,k
as follows:
T
i
= (T
i,0
T
i,1
... T
i,n1
)
= (
0

1
...
n1
)
q
00
p
0i
q
10
p
0i
q
(n1)0
p
0i
q
01
p
1i
q
11
p
1i
q
(n1)1
p
1i
.
.
.
.
.
.
.
.
.
q
0(n1)
p
(n1)i
q
1(n1)
p
(n1)i
q
(n1)(n1)
p
(n1)i
= (
0

1
...
n1
)
p
0i
p
1i
.
.
.
p
(n1)i
q
00
q
10
q
(n1)0
q
01
q
11
q
(n1)1
.
.
.
.
.
.
.
.
.
q
0(n1)
q
1(n1)
q
(n1)(n1)
D
i
Q
T
p
m
(18)
where T denotes the matrix transpose.
Observation 5.1. The components of the vector T
i
= (T
i,0
T
i,1
... T
i,n1
) can be written as a linear
combination of the components of C = (C
0
... C
n1
).
Proof: Let
M
i
D
i
Q
T
p
m (19)
where D
i
is dened in Equation (18) and
A
i
Q
p
m M
i
= Q
p
m D
i
Q
T
p
m
M
i
= P
p
m A
i
.
(20)
Then,
T
i
= M
i
= P
p
m A
i
= CA
i
. (21)
Combining this result with Equation (17) we conclude
i
=
n1
k=0
k
T
i,k
mod p i = 0, 1, ..., n 1
=
n1
k=0
k
(CA
i
)
k
mod p i = 0, 1, ..., n 1
(22)
So, instead of T
i,k
reducing to one single component of the vector C (as one might expect
from a typical convolution operation), the Pascal convolution requires a linear combination of
the components of C. Although this operation is slightly more complicated than the Fourier
approach, the identity in Equation (8) does induce a simplication.
Observation 5.2. (Symbolic Computation of Pascal Convolution)
For the 1st order case where n = p and i = 0, ..., p 1, using Equation (19) let

M
i
M
i
,
239
using Equation (18) let

D
i
D
i
and let

A
i
Q
p

M
i
. Then, for any 0 j p
m
1 where
j = j
0
p
0
+ j
1
p
1
+... + j
m1
p
m1
and A
j
= Q
p
m M
j
,
A
j
=

A
j
m1
...

A
j
1

A
j
0
(23)
where M
j
D
j
Q
T
p
m.
Proof: The statement is clearly true for the rst order case m = 1 since j = j
0
. By induction
let j = j
0
p
0
+ j
1
p
1
+... + j
m1
p
m1
and assume that
D
j
=

D
j
m1
...

D
j
1

D
j
0
where 0 j
k
p 1 for all k = 0, ..., m 1. Consider any j
= j
0
p
0
+ ... +
j
m1
p
m1
+ j
m
p
m
and apply Equation (18) along with Lucas theorem to obtain the
following intermediate result:
D
j
m

D
j
m1
...

D
j
0
=

D
j
m
D
j
=
(
j
m
0
)
(
j
m
1
)
.
.
.
(
j
m
p1
)
(
j
0
)
(
j
1
)
.
.
.
(
j
p
m
1
)
(
j
0
)
(
j
1
)
.
.
.
(
j
p
m+1
1
)
= D
j
(24)
Therefore, D
j
=

D
j
m1
...

D
j
1

D
j
0
is true. Next, successively apply the identity (AC)
(BD) = (A B)(C D) to obtain:
M
j
m1
...

M
j
1

M
j
0
= (

D
j
m1
Q
T
p
) ... (

D
j
1
Q
T
p
) (

D
j
0
Q
T
p
)
= (

D
j
m1
...

D
j
1

D
j
0
)(Q
T
p
Q
T
p
... Q
T
p
)
= D
j
Q
T
p
m
= M
j
Finally, we arrive at the desired conclusion
(

A
j
m1
...

A
j
1

A
j
0
) = (Q
p

M
j
m1
) ... (Q
p

M
j
1
) (Q
p

M
j
0
)
= (Q
p
Q
p
...Q
p
)(

M
j
m1
...

M
j
1

M
j
0
)
= Q
p
m M
j
= A
j
.
Observation 5.2 tells us that, in order to calculate T
j
= CA
j
for arbitrary n = p
m
, one need
only calculate

A
i
for i = 0, ..., p 1 and then take successive Kronecker products. The initial
set of

A
i
for i = 0, ..., p 1 can easily be calculated by referring back to Equation (20) where
A
i
= Q
p

M
i
= Q
p

D
i
Q
T
p
.
An interesting property concerning the A
i
is that the sum
p
m
1
i=0
A
i
=
p
m
1
i=0
Q
p

D
i
Q
T
p
(where the sum is taken mod p) is a matrix of ones. This follows from two observations. First,
from the denition of D
i
in Equation (18),
p
m
1
i=0
D
i
is a matrix whose (p
m
1, p
m
1) entry
is one and all other entries are zero. Second, it can also be demonstrated that the last column
of Q
p
m must be a column of ones. Therefore, Q
p
p
m
1
i=0

D
i
Q
T
p
=
p
m
1
i=0
A
i
is a matrix of ones.
Example 5.3. For p = 2, the 1st order case n = p gives i = 0, 1; hence, over GF(2),
P
p
=
1 1
0 1
, Q
p
= P
p
=
1 1
0 1
,
we calculate
A
0
= Q
p

D
0
Q
T
p
=
1 1
0 1

1 0
0 0

1 0
1 1
1 0
0 0
A
1
= Q
p

D
1
Q
T
p
=
1 1
0 1

1 0
0 1

1 0
1 1
0 1
1 1
.
From Observation 5.2, to obtain the A
j
for n = p
2
and j = 0, 1, 2, 3, one need only take successive
Kronecker products as:
A
0
=

A
0

A
0
=
1 0 0 0
0 0 0 0
0 0 0 0
0 0 0 0
, A
1
=

A
0

A
1
=
0 1 0 0
1 1 0 0
0 0 0 0
0 0 0 0
,
A
2
=

A
1

A
0
=
0 0 1 0
0 0 0 0
1 0 1 0
0 0 0 0
, A
3
=

A
1

A
1
=
0 0 0 1
0 0 1 1
0 1 0 1
1 1 1 1
.
As expected, the A
i
are symmetric matrices. Also, notice, as mentioned above, that
p
m
1
i=0
A
i
is a
matrix of ones. For the case where n = p
2
, let us now apply Observation 5.1 to calculate the Pascal
convolution of the vectors C = (C
0
, C
1
, C
2
, C
3
) and = (
0
,
1
,
2
,
3
). Using Equation (22), we
have:
0
=
0
C
0
+
1
(0) +
2
(0) +
3
(0)
1
=
0
C
1
+
1
(C
0
+ C
1
) +
2
(0) +
3
(0)
2
=
0
C
2
+
1
(0) +
2
(C
0
+ C
2
) +
3
(0)
3
=
0
C
3
+
1
(C
2
+ C
3
) +
2
(C
1
+ C
3
) +
3
(C
0
+ C
1
+ C
2
+ C
3
).
(25)
241
To close this section, we draw some immediate conclusions from Equation (25):
Because of the Kronecker product, a good deal of self-similar structure can be observed in
the resulting vector . For instance, the coefcients of the
i
can be computed by iteration
starting with the initial seed generated by

A
0
and

A
1
. As an example, the coefcient of
1
in
1
can be computed by adding the coefcient of
0
in
0
to the coefcient of
0
in
1
. The coefcients of
2
and
3
in
2
and
3
can be computed by adding the coefcients
of
0
and
1
in
0
and
1
to the coefcients of
0
and
1
in
2
and
3
, and so on.
Looking at the result columnwise, the set of coefcients associated with a given
i
appear
to be the checksums for an R(r, 2) binary Reed-Muller code ((MacWilliams & Sloane, 1977,
p.385-388), (Wicker, 1994, p.155-165)). As pointed out in the next section, although this
observation is true for the binary case, an orthogonal set of checksums for p > 2 will not
come about by this method. It is the dual of the Pascal convolution that will lead to the
decoding of GRM codes.
6. Majority logic decoding using Pascal convolution
GRM codes fall into a larger category of codes known as Euclidean geometry codes (Blahut
(2003); Lin & Costello (1983); MacWilliams & Sloane (1977); Wicker (1994)) where it is
well-known that a technique known as majority logic decoding (MLD) can be used to
recover the message vector. Based upon statements made in Section 4, it should be clear
that Pascal codes are also MLD. However, the role played by the Pascal convolution in the
decoding strategy is worthy of mention. As pointed out in the conclusions of Example 5.3, the
checksums of a majority logic decoding (MLD) scheme for GRM codes can be derived using
the dual of the convolution relation derived above. We now demonstrate this observation
more clearly.
Because of the similar forms of P
p
m and Q
p
m the dual convolution relation is easily derived
from the inverse transform. Consider the componentwise product
j
= C
j
j
of two vectors
where C = P
p
m and = P
p
m:
i
=
n1
j=0
j
q
ji
mod p i = 0, 1, ..., n 1
=
n1
j=0
(C
j
j
)q
ji
mod p
=
n1
j=0
C
j
(
n1
k=0
k
p
kj
)q
ji
mod p
=
n1
k=0
k
(
n1
j=0
C
j
p
kj
q
ji
) mod p
n1
k=0
k
s
i,k
mod p i = 0, 1, ..., n 1
(26)
where n = p
m
. Similar to Equation (18), one can also show that
s
i
= (s
i,0
s
i,1
... s
i,n1
)
= C
i
P
T
p
m
(27)
which can also be written as
s
i
= P
p
m
i
P
T
p
m (28)
where
i
is a diagonal matrix with elements (q
0i
q
1i
... q
(n1)i
) along its diagonal.
Furthermore, if we dene
B
i
P
p
m
i
P
T
p
m (29)
then results similar to Observations 5.1 and 5.2 can also be demonstrated. However, in
proving the dual of Observation 5.2 there is one difference be aware of. Since q
ji
= (1)
ij
(
i
j
),
the Kronecker product in the dual of Equation (24) will contain extra factors of (1)
ij
. To
achieve the equality
j
=

j
m1
...

j
1

j
0
where j = j
0
p
0
+ j
1
p
1
+ ... + j
m1
p
m1
the
following identity will be required:
(1)
k
= (1)
k
0
p
0
+k
1
p
1
+...+k
m1
p
m1
= (1)
k
0
((1)
p
)
k
1
((1)
p
2
)
k
2
...((1)
p
m1
)
k
m1
= (1)
m1
l=0
k
l
for any 0 k p
m
1 where we have applied a
p
= a for any a GF(p). Then, following
the proof of Observation 5.2, it is straightforward to show that for any 0 j p
m
1 where
j = j
0
p
0
+ j
1
p
1
+... + j
m1
p
m1
,
B
j
=

B
j
m1
...

B
j
1

B
j
0
(30)
where
B
j
k
= P
p

j
k
P
T
p
.
In Section 4, we explained that the form of message vectors when applying P
p
m as the
transformation where the message vector = (
0
, ...,
p
m
1
) should have all components
j
= 0 if w
p
(j) > r (see Examples 4.9 and 4.10). To see how this formulation can lead to a
decoding scheme, let us examine the case where p = 2, m = 2 and r = 1 (i.e. - a 1
st
order
binary Reed-Muller code of length 4). Consider rst using Equations (26) and (27) to calculate
Pascal convolution of the vectors = (
0
,
1
,
2
,
3
) and = (
0
,
1
,
2
,
3
):
00 :
0
=
0
C
0
+
1
(0) +
2
(0) +
3
(0)
01 :
1
=
0
(C
0
+ C
1
) +
1
C
1
+
2
(0) +
3
(0)
10 :
2
=
0
(C
0
+ C
2
) +
1
(0) +
2
C
2
+
3
(0)
11 :
3
=
0
(
3
i=0
C
i
) +
1
(C
1
+ C
3
) +
2
(C
2
+ C
3
) +
3
C
3
. (31)
243
where the binary expansion of the index has been explicitly written out at the beginning of
each row. Next, consider Equations (26) and (28) to calculate the same convolution:
00 :
0
=
0
0
+
1
(0) +
2
(0) +
3
(0)
01 :
1
=
0
1
+
1
(
0
+
1
) +
2
(0) +
3
(0)
10 :
2
=
0
2
+
1
(0) +
2
(
0
+
2
) +
3
(0)
11 :
3
=
0
3
+
1
(
2
+
3
) +
2
(
1
+
3
) +
3
(
3
i=0

i
)
.
Since, for P
2
(1, 2), = (
0
,
1
,
2
, 0), this set of equations can be simplied as
00 :
0
=
0
0
+
1
(0) +
2
(0) +
3
(0)
01 :
1
=
0
1
+
1
(
0
+
1
) +
2
(0) +
3
(0)
10 :
2
=
0
2
+
1
(0) +
2
(
0
+
2
) +
3
(0)
11 :
3
=
0
(0) +
1
2
+
2
1
+
3
(
0
+
1
+
2
)
. (32)
Equations (31) and (32) must hold for any vector . Therefore, for a specic
j
, we can equate
the coefcients of the
i
in Equation (31) with those in Equation (32). So, for example, we end
with the result that
2
= C
0
+ C
2
2
= C
1
+ C
3
and
1
= C
0
+ C
1
1
= C
2
+ C
3
.
For this rst order r = 1 code, we can generate a set of checksums using a simple algorithm.
Start at an index i of such that w
2
(i) = 1 and equate Equations (31) and (32) along a diagonal
path in order to collect all checksum equations associated associated with
i
. For example,
the bold symbols in Equation (32) generate the checksums for
1
. It turns out that these
diagonal equations actually generate what are known as the incidence vectors of the MLD
strategy (Blahut (2003); MacWilliams & Sloane (1977); Wicker (1994)).
We now provide an algorithm for GF(p) to show how the Pascal convolution approach is
equivalent to a typical MLD using nite Euclidean geometry ((Wicker, 1994, p.155-165)).
The interesting aspect of this algorithm is that the Pascal convolution generates the correct
checksums for any GF(p). Consider a P
p
(r, m) code where C = P
p
m such that
j
= 0 if
w
p
(j) > r:
(0) Let j = r.
(1) Let S
j
be the set of indices i such that w
p
(i) = j.
(2) Apply Equation (27) to calculate .
(3) Apply Equation (28) to calculate (these equations will simplify based upon which of
the
i
are zero).
(4) For each i S
j
, start at
0
associated with
i
and construct checksum equations by
equating the result in Step (2) with that of Step (3) along a diagonal path (i.e. - starting at
k=0, choose the coefcient of
k
associated with
i+k
).
(5) For i S
j
, create estimates
i
by a majority logic decision on the checksums.
(6) j = j 1. If j < 0, stop.
(7) Remove the estimated components as:
C = P
p
m
C C

C (= ( )P
p
m ).
(8) Adjust to reect the change in step (7) as follows. Construct a new vector where

i
=
i
if i S
j
and
i
= 0 otherwise. Then let
.
(9) Let C =

C and = and go to Step (1).
As with typical MLDschemes, this algorithmstarts with the highest order r to obtain estimates
of the code vector components and then successively estimates the lower order components.
Example 6.1. Let p = 3, m = 2 and r = 2. Consider decoding a P
3
(2, 2) code. From Example 4.10,
= (
0
,
1
,
2
,
3
,
4
, 0,
6
, 0, 0).
Also, we know that P
3
(2, 2) has d
min
= 3 implying that we can correct a single error. Therefore, we
expect that the MLD equations should have at least three checksums.
(0) Start with j = 2.
(1) Let S
2
= {2, 4, 6} (i.e. - i = i
0
+ i
1
p such that w
3
(i) = 2).
(2,3,4) Rather than write out the equations for
i
, we summarize by equating the results of step (2)
and step (3):
i = 2 :
2
= c
0
+ c
1
+ c
2
2
= c
3
+ c
4
+ c
5
2
= c
6
+ c
7
+ c
8
i = 4 :
4
= c
0
+2c
1
+2c
3
+ c
4
2
4
= 2c
1
+ c
2
+ c
4
+2c
5
2
4
= 2c
3
+ c
4
+ c
6
+2c
7
4
= c
4
+2c
5
+2c
7
+ c
8
i = 6 :
6
= c
0
+ c
3
+ c
6
6
= c
1
+ c
4
+ c
7
6
= c
2
+ c
5
+ c
8
245
After estimating the message components dictated by S
2
(step (5)), remove the code estimates from
C (step (7)) and begin work on S
1
where now (step(8))
i
= 0 if w
p
(i) > 1. For S
1
, we have the
checksums:
i = 1 :
1
= 2c
0
+ c
1
2
1
= c
1
+2c
2
1
= 2c
3
+ c
4
2
1
= c
4
+2c
5
1
= 2c
6
+ c
7
2
1
= c
7
+2c
8
i = 3 :
3
= 2c
0
+ c
3
3
= 2c
1
+ c
4
3
= 2c
2
+ c
5
2
3
= c
3
+2c
6
2
3
= c
4
+2c
7
2
3
= c
5
+2c
8
After estimating the message components dictated by S
1
, once again, remove the code estimates from
C and begin work on S
0
where now
i
= 0 if w
p
(i) > 0. At this stage, with all other components of
= 0 except
0
, we are left with = C (i.e. - nine estimate of the check on
0
).
7. Conclusions
When considering the design of error control codes, it is interesting to look for guiding
principles that can account for whole classes of codes. In this presentation, we have shown
how the GFFT convolution approach to Reed-Solomon codes can be extended to other classes
of codes such as generalized Reed-Muller codes.
Code Convolution Principle Decoding Strategy
Reed-Solomon GFFT-based iterative
GRM generalized iterative
Table 1. Comparison of Fourier and generalized convolution techniques.
Instead of applying a Fourier matrix to encode the message, we have applied a Pascal matrix
and extended the convolution theorem over nite elds. In doing so, we have observed
that this formulation leads to the well-known majority logic decoding algorithm. Additional
investigations have also considered codes in the context of the wavelet transform (Sakk
& Wicker (2003)). The block codes addressed in this chapter have been shown to lend
themselves to graph-based iterative decoding strategies (see Table 1). The results derived
above suggest that the generalized convolution approach is useful for understanding the
systematic introduction of redundancy for the sake of error control.
8. References
Ahmed, N., Rao, K. & Abdussattar, A. (1973). On cyclic autocorrelation and the
Walsh-Hadamard transform, IEEE Transactions on Electromagnetic Compatibility
18: 141146.
Blahut, R. (2003). Algebraic Codes for Data Transmission, Cambridge University Press.
Blahut, R. & Burrus, C. (1991). Algebraic Methods for Signal Processing and Communications
Coding, Springer.
Burrus, C., Gopinath, R. & Guo, H. (1998). Introduction to Wavelets and Wavelet Transforms,
Prentice-Hall, NJ.
Caire, G., Grossman, R. & Poor, H. (1993). Wavelet transforms and associated nite cyclic
groups, IEEE Transactions on Information Theory 39: 11571166.
Call, G. & Velleman, D. (1993). Pascals matrices, American Mathematical Monthly 100: 372376.
Dodd, M. (2003). Applications of the Discrete Fourier Transform in Information Theory and
Cryptology, PhD thesis, Royal Holloway and Bedford New College, University of
London.
Forney, G. D. (1988). Coset codes - Part II: Binary lattices and related codes, IEEE Transactions
on Information Theory 34: 11521187.
Heller, S. (1963). Inverse of triangular matrix, American Mathematical Monthly 70: 334.
Kou, Y., Lin, S. & Fossorier, M. P. C. (2001). Low-density parity-check codes based on nite
geometries: A rediscovery and new results, IEEE Transactions on Information Theory
47: 27112736.
Li, G., Li, D., Wang, Y. & Sun, W. (2010). Hybrid decoding of nite geometry low-density
parity-check codes, IET Communications 4(10): 12381246.
Lin, S. &Costello, D. (1983). Error Control Coding: Fundamentals and Applications, Prentice-Hall,
New York.
Liu, Z. & Pados, D. A. (2005). Decoding algorithm for nite-geometry LDPC codes, IEEE
Transactions on Communications 53: 415421.
MacWilliams, F. J. & Sloane, N. J. A. (1977). The Theory of Error-Correcting Codes,
North-Holland, Amsterdam.
Massey, J. L., Costello, D. J. & Justesen, J. (1973). Polynomial weights and code constructions,
IEEE Transactions on Information Theory 19: 101110.
Ngatched, T. M. N., F, T. & Bossert, M. (2009). An improved decoding algorithm for
nite-geometry LDPC codes, IEEE Transactions on Communications 57: 302306.
O.Vontobel, P., Smarandache, R., Kiyavash, N., Teutsch, J. & D.Vukobratovic (2005). On the
minimal pseudocodewords of codes from nite geometries, Proc. IEEE Int. Symp. Inf.
Theory, Adelaide, Australia.
Pusane, A. E., Smarandache, R., Vontobel, P. O. & Costello, D. J. (2011). Deriving good LDPC
convolutional codes from LDPC block codes, IEEE Transactions on Information Theory
57: 835857.
Robinson, G. (1972). Logical convolution and discrete Walsh and Fourier power spectra, IEEE
Transactions on Audio and Electroacoustics 20: 271280.
Sakk, E. (2002). Wavelet Packet Formulation of Generalized Reed Muller Codes, PhD thesis, Cornell
University, Ithaca, NY.
247
Sakk, E. & Wicker, S. (2003). Wavelet packets for error control coding, Proceedings of the SPIE
Volume 5207 (Wavelets X), San Diego, CA.
Smarandache, R., Pusane, A. E., Vontobel, P. O. & Costello, D. J. (2009). Pseudocodeword
performance analysis for LDPC convolutional codes, IEEE Transactions on Information
Theory 55: 25772598.
Strang, G. & Nuygen, T. (1996). Wavelets and Filter Banks, Wellesley-Cambridge Press,
Wellesley, MA.
Tang, H., Xu, J., Lin, S. & Abdel-Ghaffar, K. A. S. (2005). Codes on nite geometries, IEEE
Transactions on Information Theory 51: 572596.
Vandendriesscher, P. (2010). Some low-density parity-check codes derived from nite
geometries, Designs, Codes and Cryptography 54: 287297.
Wicker, S. (1994). Error Control Systems for Digital Communication and Storage, Prentice Hall.
Wicker, S. & Kim, S. (2003). Fundamentals of codes, graphs, and iterative decoding, Kluwer.
Xia, S.-T. & Fu, F.-W. (2008). Minimum pseudoweight and minimum pseudocodewords of
LDPC codes, IEEE Transactions on Information Theory 54: 480485.
Zhang, L., Huang, Q. &Lin, S. (2010). Iterative decoding of a class of cyclic codess, Information
Theory and Applications Workshop (ITA), San Diego, CA.
0
Application of the Weighted Energy Method in the
Partial Fourier Space to Linearized Viscous
Conservation Laws with Non-Convex Condition
Yoshihiro Ueda
Faculty of Maritime Sciences, Kobe University
Japan
1. Introduction
As you know, the energy method in the Fourier space is useful in deriving the decay estimates
for problems in the whole space R
n
. Recently, the author studied half space problems in R
n
+
=
R
+
R
n1
and developed the energy method in the partial Fourier space obtained by taking
the Fourier transform with respect to the tangential variable R
n1
. Then the author applied
this energy method to the half space problem for linearized viscous conservation laws with
convex condition and proved the asymptotic stability of planar stationary waves by showing
a sharp convergence rate for t (see, [14]).
In this chapter, we consider the half space problem for linearized viscous conservation laws
with non-convex condition, and derive the asymptotic stability of planar stationary waves and
the corresponding convergence rate. Our proof is based on the energy method in the partial
Fourier space with the anti-derivative method.
In this present chapter, we are concerned with the half space problem for the viscous
conservation laws:
u
t
u + f (u) = 0, (1.1)
u(0, x
, t) = u
b
, (1.2)
u(x, 0) = u
0
(x). (1.3)
Here x = (x
1
, , x
n
) is the space variable in the half space R
n
+
= R
+
R
n1
with n 2;
we sometimes write as x = (x
1
, x
) with x
1
R
+
and x
= (x
2
, , x
n
) R
n1
; u(x, t) is the
unknown function, u
0
(x) is the initial data satisfying
u
0
(x) 0 as x
1
,
and u
b
is the boundary data (assumed to be a constant) with u
b
< 0; f (u) = ( f
1
(u), , f
n
(u))
is a smooth function of u R with values in R
n
and satises
f
1
(0) = 0, f
1
(u) > f
1
(0) (= 0) (1.4)
for u [u
b
, 0). Here we note that the condition (1.4) is the necessary condition for the
existence of the planar stationary waves (for the detail, see Section 2.2). We emphasize that
10
the assumption (1.4) is weaker than the convex condition
f
1
(u) > 0 (1.5)
for u [u
b
, 0] and f
1
(0) = 0. Namely, we do not assume the convex condition for our problem
(1.1)(1.3).
For viscous conservation laws (1.1) with the convex condition (1.5), there are many results on
the asymptotic stability of nonlinear waves. First, Ilin and Oleinik in [3] studied the stability
of nonlinear waves in the one-dimensional whole space. Liu, Matsumura and Nishihara
in the paper [8] discussed the stability of stationary waves in one-dimensional half space.
More precisely, they proved the asymptotic stability of several kind of nonlinear waves
such as rarefaction waves, stationary waves, and the superposition of stationary waves and
rarefaction waves. Later, in a series of papers [57], their stability result of stationary waves in
one-space dimension was generalized to the multi-dimensional case. Kawashima, Nishibata
and Nishikawa [5] rst considered the stability of non-degenerate planar stationary waves
in two-dimensional half space and obtained the convergence rate t
1/4/2
in L
norm by
assuming that the initial perturbation is in L
2
(R
+
; L
2
(R)). Furthermore, the papers [6, 7]
studied the n-dimensional problem in the L
p
framework. In particular, the paper [7] showed
the stability of non-degenerate planar stationary waves and obtained the convergence rate
t
(n/2)(1/21/p)/2
in L
p
norm under the assumption that the initial perturbation belongs to
L
2
(R
+
; L
2
x
).
Next, we refer to viscous conservation laws with non-convex condition. Liu and Nishihara
in [9] and Nishikawa in [10] investigated the asymptotic stability of travelling waves in
the one-dimensional and multi-dimensional whole space, respectively. On the other hand,
Hashimoto and Matsumura in [1] studied the asymptotic stability of stationary waves in
the one-dimensional half space. Especially, in order to relax the convex condition, Liu and
Nishihara in [9] and Nishikawa in [10] employed the anti-derivative method and achieved the
desired result. Moreover, Hashimoto and the author in [2] used the same method to derive
the asymptotic stability of stationary waves for damped wave equations with non-convex
convection term in one-dimensional half space. Inspired by these arguments, we try to
relax the convex condition (1.5) and get the asymptotic stability of planar stationary wave
for the multi-dimensional problem (1.1)(1.3). Unfortunately, Nishikawa in the paper [10]
considered some special situation for the nonlinear term to make a good combination of the
energy method and the anti-derivative method. For the same reason, we will treat the special
situation (for the detail, see Section 3).
All these stability results mentioned above are obtained by employing the energy method in
the physical space. On the other hand, it is useful to apply the energy method in the partial
Fourier space to show sharper convergence rate. Indeed the authors paper [14] considered
our problem (1.1)(1.3) with the convex condition (1.5) and obtained the sharper convergence
rate of the planar stationary waves. We shall show the result of the paper [14] in detail.
We are interested in the asymptotic stability of one-dimensional stationary solution (x
1
)
(called planar stationary wave) for the problem (1.1)(1.3): (x
1
) is a solution to the problem
x
1
x
1
+ f
1
()
x
1
= 0, (1.6)
(0) = u
b
, (x
1
) 0 as x
1
. (1.7)
Application of the Weighted Energy Method in the Partial Fourier Space to Linearized Viscous Conservation Laws with Non-Convex Condition 3
To show the stability, it is convenient to introduce the perturbation v and write the solution u
in the form
u(x, t) = (x
1
) + v(x, t).
The original problem (1.1)(1.3) is then reduced to
v
t
v + ( f ( + v) f ()) = 0, (1.8)
v(0, x
, t) = 0, (1.9)
v(x, 0) = v
0
(x), (1.10)
where v
0
(x) = u
0
(x) (x
1
); notice that v
0
(x) 0 as x
1
.
Under the convex condition (1.5), the author in [14] showed the asymptotic stability of the
planar stationary wave (x
1
) by proving a sharp decay estimate for the perturbation v(x, t).
To this end we employed the energy method in the partial Fourier space

R
n
+
= R
+
R
n1
which is obtained by taking the Fourier transform with respect to the tangential variable x
=
(x
2
, , x
n
) R
n1
; = (
2
, ,
n
) R
n1
is the Fourier variable corresponding to x

R
n1
. For the variable x
1
R
+
in the normal direction, we use L
2
space (or weighted L
2
space). As the result, for the corresponding linearized problem with f ( + v) f () replaced
by f
()v in (1.8), we showed the following pointwise estimate with respect to R

n1
:
|Fv(, , t)|
L
2 Ce
||
2
t
|Fv
0
(, )|
L
2 , (1.11)
where C and are positive constants. Here F denotes the Fourier transform with respect to
x
R
n1
and | |
L
2 is the L
2
norm with respect to x
1
R
+
. This pointwise estimate (1.11)
enables us to get the following sharp decay estimate:
v(t)
L
2 Ct
(n1)/4
v
0
L
2
(L
1
)
, (1.12)
where
L
2 denotes the L
2
norm with respect to x = (x
1
, x
) R
n
+
,
L
2
(L
1
)
is the norm in
L
2
(R
+
; L
2
x
L
1
x
), and C is a positive constant.

Furthermore, when the planar stationary wave (x
1
) is non-degenerate, the author applied
the weighted energy method in the partial Fourier space

R
n
+
= R
+
R
n1
. Namely, we used
the weighted space L
2
with respect to x
1
R
+
. In this case, the pointwise estimate (1.11) is
improved to
|Fv(, , t)|
L
2 C(1 + t)
/2
e
||
2
t
|Fv
0
(, )|
L
2
, (1.13)
where | |
L
2
denotes the L
2

1
R
+
. Consequently, we had the decay
estimate
v(t)
L
2 C(1 + t)
/2
t
(n1)/4
v
0
L
2
(L
1
)
, (1.14)
where
L
2
(L
1
)
is the norm in L
2
(R
+
; L
2
x
L
1
x
). For the above results, we refer to the reeder

[14] in detail.
The main purpose of this chapter is to derive the sharp decay estimate (1.11)(1.14) for the
linearized problem of (1.8)(1.10) with non-convex condition (1.4), i.e.,
v
t
v + ( f
()v) = 0 (1.15)
251
Application of the Weighted Energy Method
in the Partial Fourier Space to Linearized Viscous Conservation Laws with Non-Convex Condition
with (1.9), (1.10). To overcome the difculty occured by the non-convex condition, we make
a good combination of the weighted energy method in partial Fourier space employed in [14]
and the anti-derivative method employed in [2, 9], and get the desired results. Once we obtain
the linear stability results for the problem (1.15), (1.9), (1.10), we may apply this results to the
asymptotic stability for the nonlinear problem (1.8)(1.10).
The remainder of this chapter is organized as follows. In Section 2, we introduce function
spaces and some preliminaries used in this chapter. Especially, we reformulate our problem
(1.15), (1.9), (1.10) by using the anti-derivative method in Section 2.3. In the nal section,
we treat the half space problem for the reformulated viscous conservation laws and (1.15),
(1.9), (1.10), and develop the weighted energy method in the partial Fourier space with the
anti-derivative method. In this section, we derive pointwise estimates of solutions and prove
the corresponding decay estimates.
2. Preliminaries
2.1 Notations and function spaces
Let us consider functions dened in the half space R
n
+
= R
+
R
n1
. We sometimes write the
space variable x = (x
1
, , x
n
) R
n
+
as x = (x
1
, x
) with x
1
R
+
and x
= (x
2
, , x
n
)
R
n1
. The symbols = (
x
1
, ,
x
n
) and =
n
j=1

2
x
j
denote the standard gradient and
Laplacian with respect to x = (x
1
, . . . , x
n
), respectively. The symbol
x
= (
x
2
, ,
x
n
)
denotes the gradient of the tangential direction with respect to x = (x
2
, . . . , x
n
). Thus we
have g =
n
j=1

x
j
g
j
for g = (g
1
, , g
n
), and
x
g
=
n
j=2

x
j
g
j
for g
= (g
2
, , g
n
).
Let v(x
1
, ) be the Fourier transform of v(x
1
, x
) with respect to x
R
n1
:
v(x
1
, ) = F[v(x
1
, )]() = (2)
(n1)/2
_
R
n1
v(x
1
, x
)e
ix
dx
, (2.1)
where = (
2
, ,
n
) R
n1
= (x
2
, , x
n
)
R
n1
and x
=
n
j=2
x
j
j
.
Let 1 p . We denote by L
p
x
= L
p
(R
n1
) the L
p
space with respect to x
R
n1
, with
the norm
L
p
x
. For a nonnegative integer s, we denote by H

s
x
= H
s
(R
n1
) the Sobolev
space over R
n1
with the norm
v
H
s
x
=
_ s
k=0
k
x
v
2
L
2
x
_
1/2
,
where
k
x
denotes the totality of all the k-th order derivatives with respect to x
R
n1
. Also,
we denote by L
p
(R
+
) the L
p
space with respect to x
1
R
+
, with the norm | |
L
p . For a
nonnegative integer s, we denote by H
s
(R
+
) the Sobolev space over R
+
, with the norm| |
H
s .
For R, we denote by L
2
(R
+
) the weighted L
2
space over R
+
with the norm
|v|
L
2
=
_
_

0
(1 + x
1
)
|v(x
1
)|
2
dx
1
_
1/2
.
Now we introduce function spaces over the half space R
n
+
= R
+
R
n1
. Let 1 p, q , s
be a nonnegative integer, and R. The space L
q
(L
p
) = L
q
(R
+
; L
p
x
) consists of L
q
functions
of x
1
R
+
with values in L
p
x
with respect to x
R
n1
. The norm is denoted by
L
q
(L
p
)
.
When q = p, we simply write as
L
p
= L
p
(L
p
),
L
p =
L
p
(L
p
)
.
The space H
s
(L
p
) = H
s
(R
+
; L
p
x
) consists of H
s
functions of x
1
R
+
with values in L
p
x
with
respect to x
R
n1
. The norm is denoted by
H
s
(L
p
)
. Also, L
2
(L
p
) = L
2
(R
+
; L
p
x
) denotes
the space of L
2
functions of x
1
R
+
with values in L
p
x
with respect to x
R
n1
. The norm
is denoted by
v
L
2
(L
p
)
=
_
_

0
(1 + x
1
)
v(x
1
, )
2
L
p
x
dx
1
_
1/2
.
We sometimes use
L
2
= L
2
(L
2
),
L
2
=
L
2
(L
2
)
.
L
2
(H
s
) = L
2
(R
+
; H
s
x
) denotes the space of L

2
functions of x
1
R
+
with values in H
s
x
with
respect to x
R
n1
, whose norm is given by
v
L
2
(H
s
)
=
_
_

0
v(x
1
, )
2
H
s
x
dx
1
_
1/2
=
_ s
k=0
_

0
k
x
v(x
1
, )
2
L
2
x
dx
1
_
1/2
=
_ s
k=0
k
x
v
2
L
2
_
1/2
.
By the denition (2.1) of the Fourier transform, we see that
sup
R
n1
| v(, )|
L
2 Cv
L
2
(L
1
)
(2.2)
with C = (2)
(n1)/2
. Also, it follows from the Plancherel theorem that
k
x
v
L
2
=
_
_
R
n1
||
2k
| v(, )|
2
L
2
d
_
1/2
. (2.3)
Let T > 0 and let X be a Banach space dened on the half space R
n
+
. Then C([0, T]; X) denotes
the space of continuous functions of t [0, T] with values in X.
In this paper, positive constants will be denoted by C or c.
2.2 Stationary solution
We review the results on the stationary problem (1.6)(1.7). For the details, we refer the reader
to [2, 8, 1113].
Proposition 2.1 ([8]). Assume the condition (1.4). Then f
1
(0) 0 is necessary for the existence of
solutions to the stationary problem (1.6)(1.7). Conversely, under the condition f
1
(0) 0, we have
the following existence result:
(i) Non-degenerate case where f
1
(0) < 0: In this case the stationary problem (1.6)(1.7) admits a
unique smooth solution (x
1
) with
x
1
> 0 (resp.
x
1
< 0), provided that u
b
< 0 (resp. 0 < u
b
and
f
1
(u) < f
1
(0) for 0 < u < u
b
). The solution veries
|(x
1
)| Ce
cx
1
, x
1
> 0,
253
where C and c are positive constants.
(ii) Degenerate case where f
1
(0) = 0: In this case the problem (1.6)(1.7) admits a unique smooth
solution (x
1
) if and only if u
b
< 0. The solution veries
x
1
> 0 and
|(x
1
)| C(1 + x
1
)
1/q
, x
1
> 0,
where q is the degeneracy exponent of f
1
and C is a positive constant.
In this chapter we only treat the stationary solutions (x
1
) with
x
1
> 0 and discuss
their stability; however, we must get the similar stability result of the monotone decreasing
stationary solutions by using the same argument introduced in this chapter. (We refer the
reader to [2].)
2.3 Reformulated problem
In this subsection we reformulate our problem by the anti-derivative method. To this end we
introduce a new function z(x, t) as
z(x, t) =
_

x
1
v(y, x
, t) dy. (2.4)
Here, we assume the integrability of v(x, t) over R
+
. This transformation is motivated by
the argument in Liu-Nishihara [9]. By using (2.4), we can reformulate (1.8)(1.10) in terms of
z(x, t) as
z
t
z + f
() z +
_

x
1
x
1
f
()
x
z dy = g
1
+
x
h
, (2.5)
z
x
1
(0, x
, t) = 0, (2.6)
z(x, 0) = z
0
(x), (2.7)
where z
0
(x) =
_
x
1
(u
0
(y, x
) (y))dy, f
() = ( f
2
(), , f
n
()), and g
1
,
x
h
are
nonlinear terms dened by h
= (h
2
, , h
n
) and
g
j
= f
j
( + z
x
1
) f
j
() f
j
()z
x
1
, h
j
=
_

x
1
g
j
dy.
Once we obtain the solution for the problem (2.5)(2.7), the differentiation v = z
x
1
is the
solution for (1.8)(1.10). Namely, we will apply the weighted energy method in the partial
Fourier space and try ot derive the global solution in time to the reformulated problem
(2.5)(2.7). We will discuss this reformulated problemin Section 3 to prove our main theorems.
2.4 Weight function
We introduce the weight function employed in the weighted energy method. Our weight
function is dened as
w(u) = (e
Au
+1)/f
1
(u) for u [u
b
, 0], (2.8)
where A is a positive constant determined in Lemma 2.2. This weight function is very
important to derive a priori estimate in the latter section. For this weight function, we obtain
the following lemma.
Lemma 2.2 ([2]). Suppose that f
1
(u) satises (1.4). Let w(u) be the weight function dened in (2.8).
Then there exists a positive constant such that if A , then w(u) satises the following conditions:
(i) (wf
1
)
(u) < 0 for u [u

b
, 0],
(ii) (wf
1
)
(u) < 0 for u [u

b
, 0].
(2.9)
Moreover, let be the stationary solution constructed in Proposition 2.1. Then the weight function
satises the following properties.
1
(0) < 0: The weight function w() satises
c < w() < C for [u
b
, 0].
1
(0) = 0: The weight function w() satises
c(1 + x
1
) < w() < C(1 + x
1
) for [u
b
, 0].
Here, C and c are some positive constants which independent of x
1
.
The detail of the proof is omitted here. For the details, we refer the reader to [2].
3. Asymptotic stability with convergence rates
In the nal section, we apply our weighted energy method in the partial Fourier space to
the linearized problem. We consider the linearized problem corresponding to the half space
problem (2.5)(2.7). Namely, we consider (2.5) with g
j
= 0 for j = 1, , n. For this linearized
equation, we treat the special situation that
" f
j
(u) are linear in u [u
b
, 0] for j = 2, , n."
Then our initial value problem of the linearized equation is written as
z
t
z + f
() z = 0 (3.1)
together with (2.6) and (2.7). Taking the Fourier transform with respect to x
R
n1
for the
linearized problem (3.1), (2.6), (2.7), we obtain
z
t
z
x
1
x
1
+||
2
z + f
1
() z
x
1
+ i f
() z = 0,
z
x
1
(0, , t) = 0,
z(x
1
, , 0) = z
0
(x
1
, ),
(3.2)
where = (
2
, ,
n
) R
n1
= (x
2
, , x
n
)
R
n1
, f
() = ( f
2
(), , f
n
()), and f
() =
n
j=2

j
f
j
(). This is the formulation of our
linearized problem in the partial Fourier space

R
n
+
= R
+
R
n1
.
255
Furthermore, we sometimes use the differentiated problem. We differentiate the problem(3.1),
(2.6), (2.7) with respect to x
1
. Then this yields our problem (1.15) together with (1.9) and (1.10),
and the corresponding problem in the partial Fourier space:
v
t
v
x
1
x
1
+||
2
v + ( f
1
() v)
x
1
+ i f
() v = 0,
v(0, , t) = 0,
v(x
1
, , 0) = v
0
(x
1
, ).
(3.3)
Here we note that v = z
x
1
. By applying the weighted energy method to the above problems,
we obtain the pointwise estimate of solutions.
3.1 Energy method
We apply the energy method to the problems (3.2) and (3.3) formulated in the partial Fourier
space and derive pointwise estimates of solutions to (3.2). We use L
2
space for the variable
x
1
R
+
in the normal direction. The result is given as follows.
Theorem 3.1 (Pointwise estimate). Let (x
1
) be a stationary solution with
x
1
> 0. Then the
solution to the problem (3.2) veries the following pointwise estimate.
1
(0) < 0: Suppose that z
0
(, ) H
2
(R
+
) for each R
n1
. Then
it holds
| z(, , t)|
L
2
Ce
||
2
t
| z
0
(, )|
L
2
, (3.4)
| z
x
1
(, , t)|
L
2 Ce
||
2
t
(| z
0
(, )|
L
2
+|( z
0
)
x
1
(, )|
L
2
_
, (3.5)
| z
x
1
x
1
(, , t)|
L
2 Ce
||
2
t
_
| z
0
(, )|
L
2
+|( z
0
)
x
1
(, )|
H
1
_
(3.6)
with = 0, for R
n1
and t 0, where | |
L
2
denotes the L
2

1
R
+
, and
C and are positive constants.
1
(0) = 0: Suppose that z
0
(, ) L
2
1
(R
+
) and ( z
0
)
x
1
(, ) H
1
(R
+
)
for each R
n1
. Then it holds that (3.4)(3.6) with = 1.

As a simple corollary we have the following decay estimate.
Corollary 3.2 (Decay estimate). Assume the same conditions of Proposition 3.1. Then the solution
to the problem (3.1), (2.6), (2.7) satises the following decay estimate.
1
(0) < 0: Suppose that z
0
H
2
(L
1
). Then this yields
k
x
z(t)
L
2
Ct
(n1)/4k/2
z
0
L
2
(L
1
)
, (3.7)
k
x
z
x
1
(t)
L
2 Ct
(n1)/4k/2
_
z
0
L
2
(L
1
)
+(z
0
)
x
1
L
2
(L
1
)
_
, (3.8)
k
x
z
x
1
x
1
(t)
L
2 Ct
(n1)/4k/2
_
z
0
L
2
(L
1
)
+(z
0
)
x
1
H
1
(L
1
)
_
(3.9)
with = 0, for t > 0, where k 0 is an integer and C is a positive constant.
1
(0) = 0: Suppose that z
0
L
2
1
(L
1
) and (z
0
)
x
1
H
1
(L
1
). Then this
yields that (3.7)(3.9) with = 1.
Proof of Theorem 3.1. Throughout this proof, we use the weighted L
2
norm:
|a|
L
2
w
=
_
_

0
w
_
(x
1
)
_
|a(x
1
)|
2
dx
1
_
1/2
,
where w is the weight function dened by (2.8). For this weighted norm, by using Lemma 2.2,
we see the following properties.
c| |
L
2 | |
L
2
w
C| |
L
2 for the non-degenerate case : f
(0) < 0,
c| |
L
2
1
| |
L
2
w
C| |
L
2
1
for the degenerate case : f
(0) = 0.
(3.10)
We prove (i) and (ii) in Theorem 3.1 in parallel. We rst derive (3.4). We multiply (3.2)
1
by
w()
z and take the real part, obtaining

1
2
w()

t
| z|
2
+

x
1
F
1
+D
1
= 0, (3.11)
where
D
1
= w()(| z
x
1
|
2
+||
2
| z|
2
)
1
2
(wf
1
)
()
x
1
| z|
2
,
F
1
=
1
2
(wf
1
)
()| z|
2
w()Re(
z z
x
1
),
and w is a weight function dened by (2.8). By virtue of (2.9)
2
in Lemma 2.2, we have
D
1
cw()(| z
x
1
|
2
+||
2
| z|
2
) + c
x
1
| z|
2
, (3.12)
where c is a some positive constant. Therefore, integrating (3.11) in x
1
R
+
, we get
t
| z|
2
L
2
w
+ c
1

D
1
(wf
1
)
(u
b
)| z(0, , t)|
2
0 (3.13)
with a positive constant c
1
, where
D
1
= | z
x
1
|
2
L
2
w
+||
2
| z|
2
L
2
w
+|
_
x
1
z|
2
L
2
. (3.14)
Here, by virtue of (2.9)
1
, the last term of the left-hand side of (3.13) is positive. We multiply
(3.13) by e
||
2
t
( > 0) to get
t
(e
||
2
t
| z|
2
L
2
w
) + e
||
2
t
(c
1

D
1
||
2
| z|
2
L
2
w
) 0.
(3.15)
Noting that

D
1
||
2
| z|
2
L
2
w
, we choose > 0 such that < c
1
and integrate (3.15) over [0, t].
This yields
e
||
2
t
| z(, , t)|
2
L
2
w
+
_
t
0
e
||
2

D
1
(, )d C| z
0
(, )|
2
L
2
w
, (3.16)
257
where C is a positive constant.
We next prove (3.5). Multiplying (3.3)
1
by

v and taking the real part, then we have
1
2
t
| v|
2
+

x
1
F
2
+D
2
= 0, (3.17)
where
D
2
= | v
x
1
|
2
+||
2
| v|
2
+
1
2
f
1
()
x
1
| v|
2
,
F
2
=
1
2
f
1
()| v|
2
Re(
v v
x
1
).
We integrate (3.17) in x
1
R
+
to obtain
t
| v|
2
L
2
+2

D
2
C| v|
2
L
2
, (3.18)
where

D
2
= | v
x
1
|
2
L
2
+||
2
| v|
2
L
2
and C is a positive constant. We multiply (3.18) by e
||
2
t
( > 0)
to get
t
(e
||
2
t
| v|
2
L
2
) + e
||
2
t
(2

D
2
||
2
| v|
2
L
2
) Ce
||
2
t
| v|
2
L
2
.
(3.19)
Then we choose > 0 such that < 2 and integrate (3.19) over [0, t]. This yields
e
||
2
t
| v(, , t)|
2
L
2
+
_
t
0
e
||
2

D
2
(, )d C| v
0
(, )|
2
L
2
+ C
_
t
0
e
||
2
| v(, , )|
2
L
2
d.
Noting that v = z
x
1
and | v|
L
2 C| v|
L
2
w
, we apply (3.16) to the above inequality. Then we get
e
||
2
t
| v(, , t)|
2
L
2
+
_
t
0
e
||
2

D
2
(, )d C(| z
0
(, )|
2
L
2
w
+| v
0
(, )|
2
L
2
). (3.20)
We shall show (3.6). Multiplying (3.3)
1
by
v
x
1
x
1
and taking the real part, then we have
1
2
t
| v
x
1
|
2
+

x
1
F
3
+D
3
= 0, (3.21)
where
D
3
= | v
x
1
x
1
|
2
+||
2
| v
x
1
|
2
+
3
2
f
1
()
x
1
| v
x
1
|
2
1
2
_
( f
1
f
1
)
f
1
_
()
x
1
| v|
2
1
2
i( f
1
f
()
x
1
| v|
2
,
F
3
=
1
2
f
1
()| v
x
1
|
2
+
1
2
{( f
1
f
1
)
() + i f
()}
x
1
| v|
2
Re( v
t

v
x
1
) (||
2
+ f
1
()
x
1
+ i f
())Re( v
v
x
1
).
Integrating (3.21) in x
1
R
+
, we have
t
| v
x
1
|
2
L
2
+2

D
3
C(| v|
2
H
1
+||| v|
2
L
2
), (3.22)
where

D
3
= | v
x
1
x
1
|
2
L
2
+ ||
2
| v
x
1
|
2
L
2
and C is a positive constant. We multiply (3.22) by e
||
2
t
( > 0) to get
t
(e
||
2
t
| v
x
1
|
2
L
2
) + e
||
2
t
(2

D
3
||
2
| v
x
1
|
2
L
2
) Ce
||
2
t
(| v|
2
H
1
+||| v|
2
L
2
). (3.23)
Then we choose > 0 such that < 2 and integrate (3.23) over [0, t]. This yields
e
||
2
t
| v
x
1
(, , t)|
2
L
2
+
_
t
0
e
||
2

D
3
(, )d
C|( v
0
)
x
1
(, )|
2
L
2
+ C
_
t
0
e
||
2
(| v(, , )|
2
H
1
+||
2
| v(, , )|
2
L
2
)d.
Thus, employing (3.16) and (3.20) to the above inequality, we obtain
e
||
2
t
| v
x
1
(, , t)|
2
L
2
+
_
t
0
e
||
2

D
3
(, )d C(| z
0
(, )|
2
L
2
w
+| v
0
(, )|
2
H
1
). (3.24)
Finally, we apply (3.10) to the estimates (3.16), (3.20) and (3.24). Then this gives the desired
estimates (3.4)(3.6) with = 0, 1. Hence the proof of Theorem 3.1 is completed.
Here, for later use, we derive the corresponding time weighted estimate. We multiply (3.15)
(or (3.19), (3.23)) with 0 < c
1
(or 0 < < 2) by (1 + t)
( 0) and integrate over [0, t].

Then this yields the desired estimate:
(1 + t)
e
||
2
t
| z(, , t)|
2
L
2
+
_
t
0
(1 + )
e
||
2

D
1
(, )d
C| z
0
(, )|
2
L
2
+ C
_
t
0
(1 + )
1
e
||
2
| z(, , )|
2
L
2
d,
(3.25)
(1 + t)
e
||
2
t
| v(, , t)|
2
L
2
+
_
t
0
(1 + )
e
||
2

D
2
(, )d
C| v
0
(, )|
2
L
2
+ C
_
t
0
(1 + )
e
||
2
| v(, , )|
2
L
2
d,
(3.26)
(1 + t)
e
||
2
t
| v
x
1
(, , t)|
2
L
2
+
_
t
0
(1 + )
e
||
2

D
3
(, )d
C|( v
0
)
x
1
(, )|
2
L
2
+ C
_
t
0
(1 + )
e
||
2
_
| v(, , )|
2
H
1
+||
2
| v(, , )|
2
L
2
_
d,
(3.27)
where 0, and C is a positive constant. These will be used in the next subsection.
259
Proof of Corollary 3.2. By virtue of the Plancherel theorem, we have (2.3). Substituting (3.4)
into (2.3), we obtain
k
x
z(t)
2
L
2
C
_
R
n1
||
2k
e
||
2
t
| z
0
(, )|
2
L
2
d
C sup
R
n1
| z
0
(, )|
2
L
2
_
R
n1
||
2k
e
||
2
t
d
Ct
(n1)/2k
z
0
2
L
2
(L
1
)
,
where C is a positive constant. Here we used (2.2) and the simple inequality
_
R
n1
||
2k
e
||
2
t
d Ct
(n1)/2k
with a constant C. By applying the same argument to the pointwise estimates (3.5) and (3.6) we
can derive the decay estimate (3.8) and (3.9), respectively. Thus this completes the proof.
3.2 Weighted energy method
In the last subsection, we restrict to the non-degenerate case f
1
(0) < 0 and apply the weighted
energy method to the problems (3.2) and (3.3). This yields sharp pointwise estimates of
solutions to (3.2). We use the weighted space L
2
(R
+
) ( 0) for x
1
R
+
in the normal
direction and this gives the additional decay (1 + t)
/2
. The result is stated as follows.
Theorem 3.3 (Pointwise estimate). Let f
1
(0) < 0 and let (x
1
) be a stationary solution with
x
1
> 0. Let 0 and suppose that z
0
(, ) L
2
(R
+
) and ( z
0
)
x
1
(, ) H
1
(R
+
) for each
R
n1
. Then the solution to the problem (3.2) veries the pointwise estimate
| z(, , t)|
L
2 C(1 + t)
/2
e
||
2
t
| z
0
(, )|
L
2
, (3.28)
| z
x
1
(, , t)|
L
2 C(1 + t)
/2
e
||
2
t
(| z
0
(, )|
L
2
+|( z
0
)
x
1
(, )|
L
2
_
, (3.29)
| z
x
1
x
1
(, , t)|
L
2 C(1 + t)
/2
e
||
2
t
_
| z
0
(, )|
L
2
+|( z
0
)
x
1
(, )|
H
1
_
(3.30)
for R
n1
and t 0, where the norms | |

L
2 , | |
H
1 and | |
L
2
are with respect to x

1
R
+
, and C
and are positive constants.
As an easy consequence, we have the following decay estimate.
Corollary 3.4 (Decay estimate). Assume the same conditions of Theorem 3.3. Let z
0
L
2
(L
1
) and
(z
0
)
x
1
H
1
(L
1
) for 0. Then the solution to the problem (3.1), (2.6), (2.7) satises the decay
estimate
k
x
z(t)
L
2 C(1 + t)
/2
t
(n1)/4k/2
z
0
L
2
(L
1
)
,
k
x
z
x
1
(t)
L
2 C(1 + t)
/2
t
(n1)/4k/2
_
z
0
L
2
(L
1
)
+(z
0
)
x
1
L
2
(L
1
)
_
,
k
x
z
x
1
x
1
(t)
L
2 C(1 + t)
/2
t
(n1)/4k/2
_
z
0
L
2
(L
1
)
+(z
0
)
x
1
H
1
(L
1
)
_
for t > 0, where k 0 is an integer and C is a positive constant.
The proof of Corollary 3.4 is completely same as the proof of Corollary 3.2 and omitted here.
Proof of Theorem 3.3. We use the weighted energy method to the problems (3.2) and (3.3)
formulated in the partial Fourier space. Our computation is similar to the one used in [4, 10,
12, 14] and divide into four steps.
Step 1. First, we show the following space-time weighted energy inequality:
(1 + t)
e
||
2
t
| z(, , t)|
2
L
2
+
_
t
0
(1 + )
e
||
2
_

D
(, ) + | z(, , )|
2
L
2
1
_
d
C| z
0
(, )|
2
L
2
+ C
_
t
0
(1 + )
1
e
||
2
| z(, , )|
2
L
2
d
(3.31)
for 0 and 0 , where
= | z
x
1
|
2
L
2
+||
2
| z|
2
L
2
+|
_
x
1
z|
2
L
2
,
and C and are positive constants. Notice that

D
0
coincides with

D
1
in (3.14).
To prove (3.31), we use the equality (3.11). Notice that, by virtue of (2.9)
1
in Lemma 2.2, we
have
F
1
c
2
| z|
2
C| z
x
1
|
2
(3.32)
with positive constants c
2
and C. Now we multiply (3.11) by (1 + x
1
)
(0 ) to get
1
2
t
_
(1 + x
1
)
| z|
2
_
+
_
(1 + x
1
)
F
1
_
x
1
+ (1 + x
1
)
D
1
+ (1 + x
1
)
1
(F
1
) = 0.
We integrate this equality over x
1
R
+
and use (3.12) and (3.32), obtaining
t
| z|
2
L
2
,w
+ c
1

D
+2c
2
| z|
2
L
2
1
(wf
1
)
(u
b
)| z(0, , t)|
2
C| z
x
1
|
2
L
2
1
(3.33)
for 0 , where we dene
|v|
L
2
,w
=
_
_

0
(1 + x
1
)
w
_
(x
1
)
_
|v(x
1
)|
2
dx
1
_
1/2
,
and C is a positive constant. Here, by virtue of (2.9)
1
, the last term of the left-hand side of
(3.33) is positive. We now observe that
|a|
2
L
2
1
|a|
2
L
2
+ C
|a|
2
L
2
for any > 0, where C
is a constant depending on . We apply this inequality to the term on

the right-hand side of (3.33) by taking a = z
x
1
. Noting that

D
| z
x
1
|
2
L
2
, we choose > 0 so
small that C c
1
. This yields
t
| z|
2
L
2
,w
+ c
3

D
+2c
2
| z|
2
L
2
1
C| z
x
1
|
2
L
2
(3.34)
261
for constants c
3
and C. Multiplying (3.34) by e
||
2
t
( > 0), we obtain
t
_
e
||
2
t
| z|
2
L
2
,w
_
+ e
||
2
t
_
c
3

D
||
2
| z|
2
L
2
_
+2c
2
e
||
2
t
| z|
2
L
2
1
Ce
||
2
t
| z
x
1
|
2
L
2
.
As in (3.15), we choose > 0 such that c
3
. Then we multiply the resulting inequality by
(1 + t)
( 0) and integrate over [0, t]. This yields

(1 + t)
e
||
2
t
| z(, , t)|
2
L
2
+
_
t
0
(1 + )
e
||
2
_

D
(, ) + | z(, , )|
2
L
2
1
_
d
C| z
0
(, )|
2
L
2
+ C
_
t
0
(1 + )
1
e
||
2
| z(, , )|
2
L
2
d
+ C
_
t
0
(1 + )
e
||
2
| z
x
1
(, , )|
2
L
2
d,
(3.35)
where C is a positive constant. Here the last term on the right-hand side of (3.35) is already
estimated in (3.25) because | z
x
1
|
2
L
2

D
0
=

D
1
. Therefore the proof of (3.31) is complete.
Step 2. Next we show the following estimate for 0:
(1 + t)
l
e
||
2
t
| z(, , t)|
2
L
2
l
+
_
t
0
(1 + )
l
e
||
2
_

D
l
(, ) + ( l)| z(, , )|
2
L
2
l1
_
d C| z
0
(, )|
2
L
2
(3.36)
for each integer l with 0 l [], where C and are positive constants. Note that if 0 is
an integer, then (3.36) with l = gives the desired estimate (3.28).
We prove (3.36) by induction with respect to the integer l with 0 l []. First we put = 0
and = in (3.31). This shows that (3.36) holds true for l = 0. Now, let 1 j [] (for 1)
and suppose that (3.36) holds true for l = j 1. In particular, we suppose that
_
t
0
(1 + )
j1
e
||
2
| z(, , )|
2
L
2
j
d C| z
0
(, )|
2
L
2
. (3.37)
Then we prove (3.36) for l = j. To this end, we put = j and = j in (3.31). This gives
(1 + t)
j
e
||
2
t
| z(, , t)|
2
L
2
j
+
_
t
0
(1 + )
j
e
||
2
_

D
j
(, ) + ( j)| z(, , )|
2
L
2
j1
_
d
C| z
0
(, )|
2
L
2
j
+ jC
_
t
0
(1 + )
j1
e
||
2
| z(, , )|
2
L
2
j
d C| z
0
(, )|
2
L
2
,
where we used (3.37) in the last estimate. This shows that (3.36) holds true also for l = j and
therefore the proof of (3.36) is complete.
Step 3. Next, when > 0 is not an integer, we show that
(1 + t)
e
||
2
t
| z(, , t)|
2
L
2
+
_
t
0
(1 + )
e
||
2

D
1
(, )d
C(1 + t)
| z
0
(, )|
2
L
2
(3.38)
for > , where

D
1
is dened in (3.12), and C and are positive constants. Notice that (3.38)
gives the desired estimate (3.28) even if > 0 is not an integer.
To prove (3.38), we recall the inequality (3.25) which is the same as (3.31) with = 0. We need
to estimate the second term on the right-hand side of (3.25). This can be done by applying the
technique due to Nishikawa in [10]. When > 0 is not an integer, we have from (3.36) with
l = [] that
(1 + t)
[]
e
||
2
t
| z(, , t)|
2
L
2
[]
C| z
0
(, )|
2
L
2
,
_
t
0
(1 + )
[]
e
||
2
| z(, , )|
2
L
2
[]1
C| z
0
(, )|
2
L
2
,
(3.39)
where C is a positive constant. Now, using a simple interpolation inequality |a|
L
2
|a|
L
2
1
|a|
1
L
2
(0 1) and the Hlder inequality, we see that

_
t
0
(1 + )
|a()|
2
L
2
d
_
_
t
0
(1 + )
|a()|
2
L
2
1
d
_
_
_
t
0
(1 + )
|a()|
2
L
2
d
_
1
,
(3.40)
provided that = + (1 ) with 0 1. We use (3.40) for a = e
||
2
t/2
z(x
1
, , t),
= [], = 1, = [] and the corresponding determined by = + (1 ).
Then, using (3.39), we arrive at the estimate
_
t
0
(1 + )
1
e
||
2
| z(, , )|
2
L
2
d
C| z
0
(, )|
2
L
2
_
_
t
0
(1 + )
[]
d
_
1
C(1 + t)
| z
0
(, )|
2
L
2
,
where we have used the fact that ( [] + 1)(1 ) = . Substituting this estimate into
(3.25), we get the desired estimate (3.38).
Step 4. Finally, we prove (3.29) and (3.30). Employing (3.38), we can estimate the last term
of the right-hand side of (3.26). Namely, we obtain
(1 + t)
e
||
2
t
| v(, , t)|
2
L
2
+
_
t
0
(1 + )
e
||
2

D
2
(, )d
C| v
0
(, )|
2
L
2
+ C(1 + t)
| z
0
(, )|
2
L
2
C(1 + t)
(| z
0
(, )|
2
L
2
+| v
0
(, )|
2
L
2
)
(3.41)
for > . Thus this yields (3.29).
On the other hand, by applying (3.38) and (3.41) to (3.27), we get
(1 + t)
e
||
2
t
| v
x
1
(, , t)|
2
L
2
+
_
t
0
(1 + )
e
||
2

D
3
(, )d
C(1 + t)
(| z
0
(, )|
2
L
2
+| v
0
(, )|
2
H
1
)
for > . This means that (3.30). Hence this completes the proof of Theorem 3.3.
263
4. References
[1] Hashimoto, I. & Matsumura, A. (2007). Large time behavior of solutions to an initial
boundary value problem on the half line for scalar viscous conservation law, Methods
Appl. Anal., Vol. 14, 4560.
[2] Hashimoto, I. & Ueda, Y. (2011). Anti-derivative method in the half space and
application to damped wave equations with non-convex convection, to appear in
Kyushu Journal of Mathematics.
[3] Ilin, A. M. & Oleinik, O. A. (1964). Behavior of the solution of the Cauchy problem
for certain quasilinear equations for unbounded increase of the time, Amer. Math. Soc.
Transl., Vol. 42, 1923.
[4] Kawashima, S. & Matsumura, A. (1985). Asymptotic stability of traveling wave
solutions of systems for one-dimensional gas motion, Commun. Math. Phys., Vol. 101,
97127.
[5] Kawashima, S.; Nishibata, S. & Nishikawa, M. (2003). Asymptotic stability of stationary
waves for two-dimensional viscous conservation laws in half plane, Discrete Contin.
Dyn. Syst., Suppl., 469476.
[6] Kawashima, S.; Nishibata, S. & Nishikawa, M. (2004). L
p
energy method for
multi-dimensional viscous conservation laws and application to the stability of planar
waves, J. Hyperbolic Differential Equations, Vol. 1, 581603.
[7] Kawashima, S.; Nishibata, S. & Nishikawa, M. (2004). Asymptotic stability of stationary
waves for multi-dimensional viscous conservation laws in half space, preprint.
[8] Liu T.-P.; Matsumura, A. & Nishihara, K. (1998). Behaviors of solutions for the Burgers
equation with boundary corresponding to rarefaction waves, SIAM J. Math. Anal., Vol.
29, 293308.
[9] Liu, T.-P. & Nishihara, K. (1997). Asymptotic behavior for scalar viscous conservation
laws with boundary effect, J. Differential Equations, Vol. 133, 296320.
[10] Nishikawa, M. (1998). Convergence rate to the traveling wave for viscous conservation
laws, Funkcial. Ekvac., Vol. 41, 107132.
[11] Ueda, Y. (2008). Asymptotic stability of stationary waves for damped wave equations
with a nonlinear convection term, Adv. Math. Sci. Appl., Vol. 18, No. 1, 329343.
[12] Ueda, Y.; Nakamura, T. & Kawashima, S. (2008). Stability of planar waves for damped
wave equations with nonlinear convection in multi-dimensional half space, Kinetic and
Related Models, Vol. 1, 4964.
[13] Ueda, Y.; Nakamura, T. & Kawashima, S. (2010). Stability of degenerate stationary
waves for viscous gases, Arch. Rational Mech. Anal., Vol. 198, 735762.
[14] Ueda, Y.; Nakamura, T. & Kawashima, S. (2011). Energy method in the partial Fourier
space and application to stability problems in the half space, J. Differential Equations, Vol.
250, 1169-1199.
1. Introduction
Since Bachelier, a French mathematician, rst tried to give a mathematical denition for
Brownian motion and used it to model the dynamics of stock process in 1900, nancial
mathematics has developed a lot. Black & Scholes (1973) and Merton (1973) respectively used
the geometric Brownian motion (GBM) to model the underlying assets price process so that
they opened the gate of easy ways to compute option prices, which led one of the major
breakthroughs of modern nance. Many good ideas have been proposed to model the stock
pricing processes since then. Merton (1976) rst introduced the jumps into the asset price
processes in his seminar paper. More recently, a lot of exponential Lvy models, including
Kous model, Variance Gamma (VG) model, Inverse Gaussian (IG) model, Normal Inverse
Gaussian (NIG) model, and CGMY model, etc., were proposed to add jumps in the nancial
models so that they can describe the statistical properties of nancial time series better [e.g.
see Cont & Tankov (2003) and references there in]. Also, serval stochastic volatility modes
were presented [e.g. see Heston (1993), Bates (1996), and Dufe et al (2000), etc.]. Empirical
nancial data indicate that these models are usually more consist with nancial markets.
Under assuming that the price of an underlying asset follows a GBM, Black & Scholes (1973)
showed that the value of a European option satises a boundary problem of heat equation
(Black-Scholes equation) so that they derived an explicit formula (Black-Scholes formula) for
the value. However, this nice analytic tractability in option pricing can not be carried over to
the most exponential Lvy models or the stochastic volatility modes for asset returns. Thus,
many new approaches, including efcient numerical methods, for option pricing have been
proposed. These approaches and methods can be classied into four major groups: The partial
integro-differential equation (PIDE) and various numerical methods for such equations; The
Monte Carlo methods via the stochastic simulation techniques for underlying asset price
processes; Directly numerical integration and various numerical methods via differential
integral transforms; Backward stochastic differential equation and its numerical methods.
Each of them has its advantages and disadvantages for different nancial models and specic
applications.
Since Stein & Stein (1991) rst used Fourier inversion method to nd the distribution of the
underlying asset in a stochastic volatility model, the Fourier transform methods have become
a very active eld of nancial mathematics. Heston (1993) applied the characteristic function
*
The work was supported by the research grant MYRG136(Y1-L2)-FST11-DD from the Research
Committee of University of Macau.

Fourier Transform Methods for Option Pricing
*
Deng Ding
Department of Mathematics, University of Macau, Macao,
China
11
approach to obtain an analytic representation for the valuation of European options in the
Fourier domain. Dufe et al (2000) offered a comprehensive survey that Fourier transform
methods are applicable to a wide range of stochastic processes, the class of exponential afne
diffusions [e.g. see Kwok et al (2010) and Schmelzle (2010)].
Carr & Madan (1999) pioneered the use of the fast Fourier transform (FFT) technique by
mapping the Fourier transform directly to option prices via the characteristic function. Since
then, many efcient numerical methods by using FFT techniques have been proposed,
and many authors have discussed these methods in rigorous detail. Lee (2004) extended
their method signicantly and proved an error analysis for these FFT methods. Lewis
(2001) generalized this approach to several general payoff functions via the convolution of
generalized Fourier transforms.
Recently, some newideas to improve the Fourier transformmethods have been raised. Fang &
Oosterlee (2008) proposed the COS method which is based on the Fourier and Fourier-cosine
expansion. Feng & Linetsky (2008) presented a new method which involves the relation
between Fourier transformand the Hilbert transform, and the Sinc expansion in Hardy spaces.
Lord et al (2008) extended the FFT-based methods to the CONV method, which is based on
the a quadrature technique and relies heavily on Fourier transform.
In this chapter, we demonstrate the application of Fourier transform as a very effective tool in
option pricing theory. Together with the fast Fourier transform technique and other important
properties of Fourier transform, we survey serval different methods for pricing European
options and some path dependent options under different nancial models.
This chapter is organized as follows. In the next section, the mathematical formulations that
will be used in this chapter are reviewed. They include a brief discussion on Fourier transform
with its important properties; an introduction of discrete Fourier transform and the FFT idea;
a denition of characteristic functions; and a brief review on the exponential Lvy models and
stochastic volatility models in nancial mathematics. In Section 3, several Fourier transform
methods for pricing European options are introduced. They include the Black-Scholes type
formulas, the FFT methods for signal underlying asset and for multi underlying assets, and
the Fourier expansion methods. In Section 4, three new Fourier transform methods for pricing
path dependent options are considered. They are the CONV method for pricing Bermudan
barrier option, the COS method for pricing Bermudan barrier option, and the fast Hilbert
transform method for pricing barrier option.
2. Fourier transforms and characteristic functions
2.1 Fourier transforms
Let g(x) be a piecewise continuous real function over R which satises the integrability
condition:
_

[g(x)[dx < .
The Fourier transform of g(x) is dened by
Tg(u) =
_

e
iux
g(x)dx, u R, (1)
Fourier Transform Methods for Option Pricing 3
where i =
1 is the imaginary unit. Given the function Tg(u), the function g(x) can be
recovered by the Fourier inversion formula:
g(x) =
1
2
_

e
iux
Tg(u)du, x R. (2)
Sometime it may be more convenient to consider the generalized Fourier transform on the
complex plane. Let a, b R with a < b. Assume
_

e
ax
[g(x)[dx <
and
_

e
bx
[g(x)[dx < .
Then, the generalized Fourier transform:
Tg(z) =
_

e
izx
g(x)dx, z C,
exists and is analytic for all z in the strip S =
_
z C : a < Imz < b
_
. Moreover, within this
strip the generalized Fourier transform may be inverted by integrating along a straight line
paralleled to the real axis:
g(x) =
1
2
_
iw+
iw
e
ixz
Tg(z)dz, x R, (3)
with a < w < b. Here, Im denotes taking the imaginary part of argument.
The following properties of Fourier transform are useful in this chapter.
P1. Differentiation: T
_
g
/
(x)
_
(u) = iuTg(u), where g
/
(x) is the derivative of g(x).
P2. Modulation: T
_
e
x
g(x)
_
(u) = Tg(u i), where is a constant.
P3. Convolution: T( f g)(u) = T f (u)Tg(u), where f g is the convolution of f (x) and g(x),
which is dened by
( f g)(x) =
_

f (y)g(x y)dy.
P4. Relation with Hilbert transform: T(sgn g)(u) = i1(Tg)(u), where sgn(x) is the signum
function, and 1 is the Hilbert transform, which is dened by the Cauchy principal value
integral:
1f (u) =
1
P.V.
_

f (x)
u x
dx.
P5. Parseval relation: f , g =
1
2
T f (u), Tg(u), where , is the inner product of two
square-integrable function dened by
f , g =
_

f (x)g(x)dx.
267 Fourier Transform Methods for Option Pricing
2.2 Fast Fourier transforms
Let N = 2
L
be a power of 2, and let x = (x
0
, x
1
, . . . , x
N1
)
T
be a given N-dimensional vector,
where v
T
presents the transpose of a vector v. The discrete Fourier transform of x is another
N-dimensional vector Tx = (y
0
, y
1
, . . . , y
N1
)
T
is dened by
y
p
=
N1
j=0
e
i
2
N
jp
x
j
, p = 0, 1, . . . , N1. (4)
Denote A
N
an N N matrix whose (p, j)
th
entry is given by
a
p j
= e
i
2
N
jp
, j, p = 0, 1, . . . , N1.
Then, the discrete Fourier transform of x is given by
Tx = A
N
x.
It is clear that the computation to nd Tx requires N
2
steps. However, the fast Fourier technique
(FFT) would require only
1
2
NL =
1
2
N log
2
N steps. The idea behind the FFT algorithm is to
take advantage of the periodicity property of N roots of unity. For the given N-dimensional
vector x, denote x
/
= (x
0
, x
2
, . . . , x
N2
)
T
and x
//
= (x
1
, x
3
, . . . , x
N1
)
T
two N/2-dimensional
vectors. Then, we nd
y
/
= A
M
x
/
and y
//
= A
M
x
//
,
where M = N/2, y
/
= (y
/
0
, y
/
1
, . . . , y
/
M1
)
T
and y
//
= (y
//
0
, y
//
1
, . . . , y
//
M1
)
T
, and the matrix
A
M
= [a
j k
]
MM
given by
a
p j
= e
i
2
N
jp
, j, p = 0, 1, . . . , M1.
It is easy to verify that the vector Tx, which is dened in (4), now is given by
y
p
= y
/
p
+ e
i
2
N
p
y
//
p
, p = 0, 1, . . . , M1,
y
M+p
= y
/
p
e
i
2
N
p
y
//
p
, p = 0, 1, . . . , M1.
Instead of performing the matrix-vector multiplication A
N
x, we now only need to perform
two matrix-vector multiplications A
M
x
/
and A
M
x
//
so that the number of operations is
reduced from N
2
to 2(N/2)
2
= N
2
/2. The same procedure of reducing the length of the
vector by half can be applied repeatedly. Using this FFT algorithm, the total number of
operations is reduced from O(N
2
) to O(N log
2
N). The similar FFT algorithm also can be
used to calculate the discrete Fourier inversion transform:
x
p
=
N1
j=0
e
i
2
N
jp
y
j
, p = 0, 1, . . . , N1.
2.3 Characteristic functions
Let X be a random variable having the density function f (x). Then, the characteristic function
of X is dened by
X
(u) = E
_
e
iuX
=
_

e
iux
f (x)dx, u R,
i.e. it is the Fourier transform of the density f (X). Here E[] denotes the mathematical
expectation. The characteristic functions are uniformly continuous and non-negative denite.
Also they possess the following properties:
P6. If X and Y are independent, then
X+Y
(u) =
X
(u)
Y
(u).
P7. If a, b R and Y = aX + b, then
Y
(u) = e
ibu
X
(au).
P8. X and Y have the same distribution function if and only if they have the same characteristic
function.
2.4 Exponential Lvy models
An adapted stochastic process X
t
with X
0
= 0 is called a Lvy process if it has independent and
stationary increments, and it is continuous in probability. Moreover, every Lvy process has
a right continuous with left limits (cdld) version which is also a Lvy process. We always
work with this cdlg version.
Let X
t
be a Lvy process, and let B
0
be the set family of all Borel sets U R whose closure

U
does not contain 0. Set
_
(0, t], U
_
=

0<st
1
U
_
X
s
_
, t > 0, U B
0
,
and denote
_
0, U
_
0. Then (dt, dx) is a -nite random measure on B(R
+
) B
0
,
which is called the jump measure of X
t
. Set (U) = E
_
_
[0, 1], U
_
for any U B
0
. It is
clear that (dx) is a -nite measure on B
0
, which is called the Lvy measure of X
t
. Then, the
characteristic function of Lvy process X
t
has the Lvy-Khintchine representation:
t
(u) = e
t(u)
, u R, (5)
where the characteristic exponent function:
(u) = iu
1
2
2
u
2
+
_
R
_
e
iux
1 iux1
[x[1
(x)
_
(dx). (6)
Here the triple (,
2
, (dx)) is called the Lvy triple of X
t
. The proof of this Lvy-Khintchine
decomposition is based on the famous Lvy-It decomposition, and the detail can be found in
Cont & Tankov (2003). Also a Lvy process is a strong Markov process, and a semimartingale.
Furthermore, under the condition:
_
[x[1
e
x
(dx) < , the exponential e
X
t
is a martingale if
and only if
+
1
2
2
+
_
R
_
e
x
1 x1
[x[1
(x)
_
(dx) = 0.
To ensure positivity as well as the independence and stationarity of log-returns, in nancial
mathematics, the price process of the underlying asset is usually modeled as the exponential
of a Lvy process X
t
:
S
t
= S
0
e
X
t
, t 0. (7)
There are many models in the nancing modeling literature simply correspond to different
choices for
2
and (dx) to the Lvy process X
t
.
The cases that
2
,= 0 and (dx) is a nite measure, i.e. X
t
is a Lvy jump diffusion:
X
t
= t + B
t
+
N
t
k=1
k
, t 0,
where B
t
is a standard Brownian motion,
k
is a sequence of independent and identically
distributed random variables with common density f
(x), and N
t
is a Poisson process of
intensity , such that B
t
,
k
and N
t
are mutually independent. In these models, the Lvy
triple is given by
= +
_
[x[<1
x f
(x)dx,
2
and (dx) = f
(x)dx.
These models explain the jump part as the market responses to outside news: good news and
bad news. The news arrives according to the Poisson process N
t
, and the price changes in
response according to the jump size
k
.
The earliest of these models is due to Merton (1976). In this model,
k
s are normally
distributed, with mean
J
and standard deviation
J
, so that the characteristic function
of X
t
is given by
t
(u) = exp
_
itu
1
2
2
tu
2
+ t
_
e
i
J
u
1
2
2
J
u
2
1
_
_
.
The second one belongs to Kou (2002). In this model,
k
s are non-symmetric double
exponentially distributed, and the characteristic function of X
t
is given by
t
(u) = exp
_
itu
1
2
2
tu
2
+ t
_
1
2
1 +
2
u
2
e
iu
1
__
,
where and are parameters.
The cases that
2
= 0 and (R 0) = , i.e., the models have innite activity in jumps. We
only list some examples here.
The VG model, which is proposed by Madan et al (1998): X
t
is a variance gamma process
with parameters , and , and its characteristic function is given by:
t
(u) =
_
1 iu +
1
2
2
u
2
_
t/
.
The NIG model, which is presented by Barndorff-Nielsen (1997): X
t
is a normal inverse
Gaussian process with parameters , and , and its characteristic function is given by:
t
(u) = exp
_
t
_
_
2
( + iu)
2
2

2
__
.
The CGMY model, which is dened by Carr et al (2002): X
t
is a CGMY process whose
characteristic function is given by
t
(u) = exp
_
tC(Y)
_
(Miu)
Y
M
Y
+ (G + iu)
Y
G
Y
_
,
where the parameters C, G and M are nonnegative, Y < 2, and (x) is the gamma function.
The Finite Moment Log Stable model, which is considered by Carr & Wu (2003): X
t
is a
nite moment log stable process whose characteristic function is given by
t
(u) = exp
_
itu (iu)
t sec

2
_
,
where , and are parameters.
2.5 Stochastic volatility models
Although the class of Lvy processes is quite rich, it is sometime insufcient for multi-period
nancial modeling. Several models combining jumps and stochastic volatility have been
appeared in the literature. We only list two such models here.
The Hestons stochastic volatility model [Heston (1993)]: In this model, the price process
S
t
is dened by the system of stochastic differential equations (SDEs):
_
dS
t
= S
t
dt +
V
t
S
t
dB
S
t
,
dV
t
= ( V
t
)dt +
V
t
dB
V
t
,
with the initial values S
0
= s
0
and V
0
= v
0
, where B
S
t
and B
V
t
are two standard Brownian
motions with correlation (i.e. B
S
, B
V
t
= t), is the mean reversion, is the long-run
variance level, and is the volatility-of-volatility parameter. Although S
t
is no longer a
Lvy process, under the risk neutral probability the characteristic function of the log-price
process X
t
= log S
t
is known in closed form:
t
(u) =
exp
_
t(iu)
2
+ irtu + ix
0
u
_
_
cosh
t
2
+
iu
sinh
t
2
_
2/
2
exp
_
(u
2
+ iu)v
0
coth
t
2
+ iu
_
,
where x
0
= log s
0
, r is the interest rate, and =
_
2
(u
2
+ iu) + ( iu)
2
.
The Bastes stochastic volatility model [Bates (1996)]: In this model, the price process S
t
and the variance process V
t
are given by the system of SDEs:
_
dS
t
= S
t
dt +
V
t
S
t
dB
S
t
+ S
t
dZ
t
,
dV
t
= ( V
t
)dt +
V
t
dB
V
t
,
where B
S
t
and B
V
t
are two standard Brownian motions with correlation , and Z
t
is a
compound Poisson process with intensity and log-normal distribution of jump sizes such
that if k is its jump size then
log(1 + k) N
_
log(1 +

k)
1
2
2
,
2
_
.
Similarly to the Hestons model, the characteristic function of X
t
= log S
t
is known in
closed form:
t
(u) =
D
t
(u)
J
t
(u),
where the characteristic function of the diffusion part:
D
t
(u) =
exp
_
t(iu)
2
+ i(r
k)tu + ix
0
u
_
_
cosh
t
2
+
iu
sinh
t
2
_
2/
2
exp
_
(u
2
+ iu)v
0
coth
t
2
+ iu
_
,
and the characteristic function of the jump part:
J
t
(u) = exp
_
t
_
e
1
2
2
u
2
+i
_
log(1+
k)
1
2
2
_
u
1
__
.
3. Fourier transform methods for pricing European options
3.1 Black-Scholes type formulas
Gurland (1948) used the Convolution property P3 to derive an formula for calculation of the
distribution function F(x) from the characteristic function (u) :
F(x) =
1
2
+
1
2
_

0
e
iux
(u) e
iux
(u)
iu
du.
Shephard (1991) gave a simple proof of this formula. Fromthis formula by a simple calculation
we can get an important formula for Fourier inversion method:
P
_
X > x
_
= 1 F(x) =
1
2
+
1
_

0
Re
_
e
iux
(u)
iu
_
du. (8)
Here Re denotes taking the real part of argument. Inside the eld of nance, this Fourier
inversion method was rst considered by Stein & Stein (1991) to nd the distribution of the
underlying asset price in a stochastic volatility model. Heston (1993) applied the formula (8),
to obtain a Black-Scholes type formula for pricing a European call option in the stochastic
volatility model.
In fact, with help of the risk-neutral valuation, the price of a European call with spot price S
0
and strike price K is given by
V(S
0
, T; K) = e
rT
E
_
(S
T
K)
+
= e
rT
_

(e
x
e
k
)
+
f
T
(x)dx
=
_

k
e
rT+x
f
T
(x)dx e
rT
K
_

k
f
T
(x)dx = I
1
e
rT
KI
2
,
where r is the interest rate, T is the maturity, k = log K, and f
T
(x) is the density function of
X
T
= log S
T
. It is clear that the second integral is the probability:
2
= P(X
t
> k). By the
Fourier inversion formula (8) we have
2
=
1
2
+
1
_

0
Re
_
e
iuk
T
(u)
iu
_
du,
where
T
(u) is the characteristic function of X
T
. By a change of probability measure, we can
verify that the second integral is give by I
1
= S
0
1
, where
1
=
1
2
+
1
_

0
Re
_
e
iuk
T
(u i)
iu
T
(i)
_
du,
Thus, the price of the European call now is given by a Black-Scholes type formula:
V(S
0
, T; K) = S
0
1
Ke
r(Tt)
2
. (9)
Beginning with Hestons work, many authors used the Fourier inversion methods to solve
advanced valuation problems [e.g. see Bates (1996), Schmelzle (2010), and references therein].
As mentioned by Carr & Madan (1999), the numerical integrals based on these methods are
generally much faster than nite difference solutions to PDEs or PIDEs. Unfortunately, the
FFT cannot be used to evaluate these integrals, since the integrands are singular at the required
evaluation point u = 0.
Recently, several authors applied the Property P5 (Parseval relation) to derive the Fourier
inversion formulas in option pricing [e.g. see Dufresne et al (2009)]. We have well known that,
under the risk-neutral probability, the option price V(S
0
, T) with the terminal payoff H(S
T
) is
given by
V(S
0
, T) = e
rT
E
_
H(S
T
)
= e
rT
_

H(x)p
T
(x)dx = e
rT
H(x), p
T
(x),
where p
T
(x) is the density function of S
T
. By the Parseval relation we have
V(S
0
, T) =
e
rT
2
TH(u), Tp
T
(u).
3.2 Fast Fourier transform methods - single asset
1. The method of Carr and Madan. Carr & Madan (1999) developed a different method
designed to use the fast Fourier transform to price options. They introduced a new
technique to calculate the Fourier transform of a modied call option price with respect to
the logarithmic strike price so that the fast Fourier transform can be applied to calculate the
integrals.
Consider a European call with the maturity T and the strike price K, which is written on a
stock whose price process is S
t
= e
rt+X
t
, under a risk-neutral probability. Let f
T
(x) be the
density of X
T
. Consider the price of a call option:
V
T
(k) = e
rT
E
__
e
X
T
e
k
_
+
= e
rT
_

k
(e
x
e
k
) f
T
(x)dx, (10)
where k = ln K is the log strike price. Note that V
T
(k) S
0
= 1 as k , and the function
V
T
(k) is not square-integrable. Thus, we cannot express the Fourier transform in strike in
terms of the characteristic function
T
(u) of X
T
and then nd a range of strikes by Fourier
inversion.
To obtain a square-integrable function we consider the modied call price:
V
T
(k : ) = e
k
V
T
(k)
for some > 0, which is chosen to improve the integrability. Carr & Madan (1999) showed
that a sufcient condition for square-integrability of

V(k) is given by
_

f
T
(x)e
(1+)x
dx < .
Consider the Fourier transform of

V(k : ):
T
(v : ) = T

V
T
(k : )(v) =
_

e
ivk

V
T
(k : )dk.
By (10) and the denition of characteristic functions we have
T
(v : ) = e
rT
_

f
T
(x)dx
_
x
_
e
x+k
e
(1+)k
_
e
ivk
dk
=
e
rT
T
_
v i( + 1)
_
2
+ v
2
+ i(2 + 1)v
, (11)
where
T
T
. Using the Fourier inverse transform we get
V
T
(k) = e
k

V(k : ) =
e
k
2
_

e
ivk
T
(v : )dv.
Note that
T
(v : ) is odd in its imaginary part and even in its real part. Since V(k) is real we
have
V
T
(k) =
e
k
_

0
e
ivk
T
(v : )dv. (12)
Now this integral can be evaluated by the numerical approximation using the trapezoidal
rule:
V
T
(k
p
)
e
k
p
2
N1
j=0
e
iv
j
k
p
T
(v
j
: )v, p = 0, 1, . . . , N 1.
to apply the FFT algorithm, we set N as a power of 2, and dene the grid points:
_
k
p
=
1
2
Nk + pk, p = 0, 1, . . . , N1,
v
j
=
1
2
Nu + jv, j = 0, 1, . . . , N1,
with the step sizes k and v, which satisfy the Nyquist relation: kv = 2/N. Then, the
numerical approximation can expressed as
V
T
(k
p
)
e
k
p
2
N1
j=0
(1)
j+p
e
i
2
N
jp
T
(v
j
: )v, p = 0, 1, . . . , N1,
which can be efciently computed by using FFT algorithm.
Many authors have discussed this Carr and Madans pricing method in rigorous detail. Lee
(2004) extended this method signicantly and proved an error analysis for this FFT method.
On the other hand, numerical experiments show that this FFT method has a quite large error
when the strike price K is small, and its stability is very dependent on the choice of the
damping exponential factor . Ding & U (2010) presented a modied FFT-based method to
overcome these disadvantages.
2. The method of Lewis. Lewis (2001) introduced an option pricing method which generalizes
previous work on Fourier transform methods. The main idea of his method is to express the
option price via the convolution of generalized Fourier transforms, and then, to apply the
property P5 (Parseval relation) to obtain the generalized Fourier transform of option price.
Consider a European type option whose payoff is H(S
T
) at maturity T, where S
t
= S
0
e
rt+X
t
is the stock price process under the risk neutral probability. To proceed, we assume that X
t
is
a Lvy process having the analytic characteristic function
T
(z), which is regular in the strip
S
X
=
_
z : a < Imz < b
_
, and h(x) = H(e
x+rT
) is Fourier integrable in some strip S
h
such
that
S
V
=

S
X
S
h
,= ,
where

S
X
is the complex conjugate set of S
X
.
Let s = log S
0
denote the logarithm of current stock value. Then, the option price is given by
V(s) = e
rT
E
_
H
_
e
s+rT+X
T
_
_
= e
rT
_

H(e
s+rT+x
) f
T
(x)dx,
where f
T
(x) is density function of X
T
. Under the assumption we can compute the generalized
Fourier transform of V(s) by the Parseval relation:
_

e
izs
V(s)ds = e
rT
_

f
T
(x)dx
_

e
izs
H(e
s+rT+x
)ds
= e
rT
_

e
izx
f
T
(x)dx
_

e
izy
H(e
y+rT
)dy,
for all z S. Finally we obtain
TV(z) = e
rT
T
(z)Th(z), z S
V
.
Option prices can now be given by the generalized Fourier inversion formula (3):
V(s) =
e
rT
2
_
iw+
iw
T
(z)Th(z)dz,
with a < w < b, for a range of initial values s.
For a European call option, the function h(x) = (e
rT+x
e
k
)
+
, is Fourier integrable in the
region Imz > 1, where its generalized Fourier transform can be computed explicitly:
Th(z) =
_

e
ixz
_
e
x+rT
e
k
_
+
dx =
e
k+iz(krT)
iz(iz + 1)
.
Hence, the generalized Fourier transform of call option price takes the form:
TV(z) =
T
(z)
e
(1+iz)(krT)
iz(iz + 1)
, 1 < Imz < 1 + .
and hence, the option price is given by
V(s) =
e
ws+(1w)(krT)
2
_

T
(z)e
iu(krTs)
(iu w)(1 + iu w)
du,
for some 1 < w < 1 + . The integral in this formula can be approximated by using the FFT
algorithm as the formula (12). However, the choice of w is a delicate issue because choosing
big w leads to slower decay rates at innity and bigger truncation errors, and while w is close
to one the denominator diverges and the discretization error becomes to large (see Chapter 11
in Cont & Tankov (2003)). On the other hand, Lewis (2001) also listed the generalized Fourier
transforms Th(z) and the strip S
h
for various claims, for instance, the put option, the covered
call or the cash-secured put, etc.
3.3 Fast Fourier transform methods - multi assets
1. The most direct extension of the Carr and Madan method to the multi assets model is to
price the correlation option, whose payoff at the maturity T is dened by
H(S
1T
, S
2T
) = (S
1T
K
1
)
+
(S
2T
K
2
)
+
,
where K
1
and K
2
are strike prices [see, e.g. Kwok et al (2010)]. Dene X
it
= log S
it
and
k
i
= log K
i
, i = 1, 2, and let f
T
(x
1
, x
2
) be the joint density of X
1T
and X
2T
under the risk neutral
probability. Then, the characteristic function of X
1T
and X
2T
is dened by the following
two-dimensional Fourier transform:
T
(u
1
, u
2
) =
_

e
i(u
1
x
1
+u
2
x
2
)
f
T
(x
1
, x
2
)dx
1
dx
2
.
Consider the price of the correlation option:
V
T
(k
1
, k
2
) = e
rT
_

k
2
_

k
1
(e
x
1
e
k
1
)(e
x
2
e
k
2
) f
T
(x
1
, x
2
)dx
1
dx
2
.
Following the Carr and Madam method we consider the modied call price:
V
T
(k
1
, k
2
:
1
,
2
) = e
1
k
2
+
2
k
2
V
T
(k
1
, k
2
),
for some parameters
1
> 0 and
2
> 0, which are chosen such that this modied price
is square integrable for negative value of k
1
and k
2
. Then, by a direct integral the Fourier
transform of

V(k
1
, k
2
) is given by
T
(v
1
, v
2
) = T

V
T
(k
1
, k
2
:
1
,
2
)
=
e
rT
T
_
v
1
i(
1
+ 1), v
2
i(
2
+ 1)
_
(
1
+ iv
1
)(
1
+ 1 + iv
1
)(
2
+ iv
2
)(
2
+ 1 + iv
2
)
.
Applying the Fourier inversion on
T
(v
1
, v
2
) and using the numerical approximation of
two-dimensional Fourier inversion integral we have
V
T
(k
1
p
, k
2
q
) =
e
1
k
1
p
2
k
2
q
(2)
2
_

e
i(v
1
k
1
p
+v
2
k
2
q
)
T
(v
1
, v
2
)dv
1
dv
2
1
k
1
p
2
k
2
q
(2)
2
N1
m=0
N1
n=0
e
i(v
1
m
k
1
p
+v
2
n
k
2
q
)
T
(v
1
m
, v
2
n
)v
1
v
2
,
for p, q = 0, 1, . . . , N1. To apply the two-dimensional of the FFT algorithm we set
v
1
m
=
_
m
N
2
_
v
1
, v
2
n
=
_
n
N
2
_
v
2
, m, n = 0, 1, . . . , N1,
and
k
1
p
=
_
p
N
2
_
k
1
, k
2
q
=
_
q
N
2
_
k
2
, p, q = 0, 1, . . . , N1,
where N is a power of 2, and the step sizes v
1
, v
2
, k
1
, and k
2
observe the Nyquist
relations: v
1
k
1
= v
2
k
2
= 2/N. Dempster & Hong (2002) shown that the numerical
approximation now is given by
V
T
(k
1
p
, k
2
q
)
e
1
k
1
p
2
k
2
q
(2)
2

T
(k
1
p
, k
2
q
),
for k
1
p
,= k
2
q
, where
T
(k
1
p
, k
2
q
) = (1)
p+q
N1
m=0
N1
n=0
e
i
2
N
(mp+nq)
_
(1)
m+n
T
(v
1
m
, v
2
n
)v
1
v
2
_
,
which can be computed via the FFT algorithm.
There are some other types of terminal payoff functions that admit analytic representation of
the Fourier transform of the damped option price [see e.g. Eberlein et al (2010), Lee (2004),
and references therein]. However, to derive the FFT pricing algorithm for the spread option:
H(S
1T
, S
2T
) = (S
1T
S
2T
K)
+
, (13)
Dempster & Hong (2002) approximated the exercise region of H(S
1T
, S
2T
) by a combination
of rectangular strips.
2. Hurd & Zhou (2010) proposed an alternative approach to pricing the European spread
option (13) under the three-factor SV model and exponential Lvy models. Let h(x
1
, x
2
) be
the terminal spread option payoff with the strike price K = 1, i.e.
h(x
1
, x
2
) = (e
x
1
e
x
2
1)
+
. (14)
And denote (z) the complex gamma function dened by
(z) =
_

0
e
t
t
z1
dt, Rez > 0.
Then, the method of Hurd and Zhou mainly relies on the following Fourier representation of
the payoff function h(x
1
, x
2
):
h(x
1
, x
2
) =
1
(2)
2
_
i
2
+
i
2
_
i
1
+
i
1
e
i(u
1
x
1
+u
2
x
2
)
h(u
1
, u
2
)du
1
du
2
, (15)
where
1
and
2
are any two given real numbers with
2
> 0 and
1
+
2
< 1, and
h(u
1
, u
2
) =
_
i(u
1
+ u
2
) 1
_
(iu
2
)
(iu
1
+ 1)
.
The proof of this formula can be found in the appendix of Hurd & Zhou (2010). Let X
1t
=
log S
1t
and X
2t
= log S
2t
with X
10
= x
1
and X
20
= x
2
, and let
T
(u
1
, u
2
) be the characteristic
function of X
1T
x
1
and X
2T
x
2
. Then, we have
E
_
e
i(u
1
X
1T
+u
2
X
2T
)
= e
i(u
1
x
1
+u
2
x
2
)
T
(u
1
, u
2
).
Now, using the formula (16), the price of the spread option (14) is expressed as an explicit
two-dimensional Fourier inversion transform:
V
T
(x
1
, x
2
) = e
rT
E
__
e
X
1T
e
X
2T
1
_
+
= e
rT
E
_
1
(2)
2
_
i
2
+
i
2
_
i
1
+
i
1
e
i(u
1
X
1T
+u
2
X
2T
)
h(u
1
, u
2
)du
1
du
2
_
=
e
rT
(2)
2
_
i
2
+
i
2
_
i
1
+
i
1
E
_
e
i(u
1
X
1T
+u
2
X
2T
)
h(u
1
, u
2
)du
1
du
2
=
e
rT
(2)
2
_
i
2
+
i
2
_
i
1
+
i
1
e
i(u
1
x
1
+u
2
x
2
)
T
(u
1
, u
2
)
h(u
1
, u
2
)du
1
du
2
.
This two-dimensional Fourier inversion integral can be numerically computed by the usual
FFT calculations. Since this approach does not require an approximation of the exercise region,
it is considered to be more computationally efcient.
3.4 Fourier expansion methods
Fang & Oosterlee (2008) proposed a new numerical method for pricing European options,
which is called the COS method. This method is based on the Fourier transform and the
Fourier-cosine expansion, and is shown to have the exponential convergence rate and linear
computational complexity. And now, this novel method has been considered by many authors
to price various options [see e.g. Ding et al (2011b),Fang & Oosterlee (2009), and the references
therein].
We recall that the price S
t
of underlying asset follows an exponential Lvy model. Let X
t
=
log(S
t
/K), where K is the strike price. Consider the price of European call in the form:
V
0
(K) = e
rT
E
_
(S
T
K)
+
= e
rT
K
_

0
(e
x
1) f
T
(x)dx, (16)
where f
T
(x) is the probability density of X
T
under the risk-neutral probability. Let
T
(u) be
the characteristic function of f
T
(x). The main ideas of the COS method are to choose two
numbers a and b such that the truncated integral approximates the innite integral very well,
i.e.,

T
(u) =
_
b
a
e
iux
f
T
(x)dx
_

e
iux
f
T
(x)dx =
T
(u), (17)
and to consider the Fourier-cosine expansion of f
T
(x) in [a, b]:
f
T
(x) =
1
2
A
0
+
n=1
A
n
cos
_
n
x a
b a
_
, x [a, b], (18)
where
A
n
=
2
b a
_
b
a
p
T
(x) cos
_
n
x a
b a
_
dx, n = 0, 1, 2, . . . .
From (17) we have
A
n
=
2
b a
Re
_

T
_
n
b a
_
exp
_
i
na
b a
__
, n = 0, 1, 2, . . . .
Then, we get an approximation of f
T
(x) in (18) by
f
T
(x)
1
2
A
0
+
N1
n=1
A
n
cos
_
n
x a
b a
_
, x [a, b], (19)
where
A
n
=
2
b a
Re
_
T
_
n
b a
_
exp
_
i
na
b a
__
A
n
, n = 0, 1, . . . , N1
Now, substituting (19) into (16), we obtain an approximation of the option price:
V
0
(K) Ke
rT
_
1
2
A
0
_
0
(0, b)
0
(0, b)
_
+
N1
n=1
A
n
_
n
(0, b)
n
(0, b)
_
_
, (20)
where
n
(0, b) and
n
(0, b) are two integrals given by:
n
(c, d) =
_
d
c
e
x
cos
_
n
x a
b a
_
dx and
n
(c, d) =
_
d
c
cos
_
n
x a
b a
_
dx,
for any [c, d] [a, b], which are analytically given by
n
(c, d) =
1
1 +
_
n
ba
_
2
_
cos
_
n
d a
b a
_
e
d
cos
_
n
c a
b a
_
e
c
+
n
b a
sin
_
n
d a
b a
_
e
d
n
b a
sin
_
n
c a
b a
_
e
c
_
,
n
(c, d) =
_
sin
_
n
d a
b a
_
sin
_
n
c a
b a
__
b a
n
n ,= 0,
(d c) n = 0
.
Fang & Oosterlee (2008) also showed that, in most cases, the convergence rate of the COS
formula (20) is exponential and the computational complexity is linear. They also discussed
the truncation range for COS method, and gave a general formula to determine the interval of
integration [a, b] in that paper, which is given by
[a, b] =
_
c
1
_
c
2
+
c
4
, c
1
+
_
c
2
+
c
4
_
, (21)
where c
1
, c
2
, and c
4
are the rst, second, and fourth cumulates of X
t
= log(S
t
/K). Also,
the constant depends on the tolerance level in the approximation (17), and usually we
choose = 10. Meanwhile, Ding & U (2011) respectively applied the Fourier-sine and Fourier
expansions to substitute the Fourier-cosine expansion in (18). A comprehensive analysis with
numerical comparisons for these different methods is also given in their paper.
4. Fourier transform methods for pricing path dependent options
4.1 The CONV method for pricing Bermudan barrier options
Pricing Bermudan or barrier options is much harder than pricing European options. Because
these options are depended on paths of the price process for the underlying assets. Recently,
some new numerical integration methods based on Fourier transforms are proposed. Lord et
al (2008) proposed an efcient and accurate FFT-based method, called the CONV method, to
price Bermudan options under exponential Lvy models.
In the following we apply the CONV method to price a Bermudan barrier option in which the
monitored dates may be many times more than the exercise dates. Denote
G(S) =
_
(S K)
+
, for call option,
(K S)
+
, for put option,
where K is the strike price, S is the spot price of the underlying asset. Let
_
T =
_
t
mL+l
: m = 0, 1, . . . , M, l = 0, 1, . . . , L1
_
,
T
e
=
_
t
mL
: m = 1, . . . , M1
_
T,
(22)
be the set of pre-specied monitored dates and the set of pre-specied exercise dates,
respectively, before maturity T, where
0 = t
0
< t
1
< < t
ML
= T with t = t
k
t
k1
= T/(ML).
Consider a discrete American barrier option, which is monitored at every t
k
T and can be
exercised at each t
k
T
e
, namely Bermudan barrier option, whose payoff is given by
G(S
t
k
)1
S
t
k
<H
+ R
0
1
S
t
k
H
, t
k
T.
Here S
t
k
is the price of the underlying asset at time t
k
T, H > K is the constant barrier and
R
0
is the contractual rebate. That is, this Bermudan barrier option is an up-and-out barrier
option that cease to exist if the asset price S
t
k
hits the barrier level H at one time t
k
T, and it
can also be exercised at any time t
k
T
e
before maturity T.
Denote V(S, t
k
) the value of this Bermudan barrier option at time t
k
and the spot price
S
t
k
= S. With help of the risk-neutral valuation formula, this price process can be computed
recursively by the following backward induction:
V(S, t
ML
) = G(S)1
S<H
+ R
0
1
SH
,
C(S, t
k
) = E
_
e
rt
V(S
t
k+1
, t
k+1
) [ S
t
k
= S
, t
k
T,
V(S, t
k
) = C(S, t
k
)1
S<H
+ e
r(Tt
k
)
R
0
1
SH
, t
k
T T
e
,
V(S, t
k
) = max
_
G(S), C(S, t
k
)
_
1
S<H
+ e
r(Tt
k
)
R
0
1
SH
, t
k
T
e
,
(23)
in specialty, the initial price is given
V(S, t
0
) = C(S, t
0
) = E
_
e
rt
V(S, t
1
) [ S
t
0
= S
.
Here r > 0 is the interest rate, and E[ [ ] is the conditional expectation under the risk-neutral
probability.
Assume that the price process of the underlying asset is given by
S
t
= S
0
e
X
t
, t 0,
where X
t
is a Lvy process and S
0
is the initial price. Let f ( [ x) be the condition density of
X
t
k+1
given X
t
k
= x for t
k
T. Set
g(x) =
_
(S
0
e
x
K)
+
, for a call option,
(K S
0
e
x
)
+
, for a put option.
Then the backward induction (23) can be rewritten by
v(x, t
ML
) = g(x)1
x<h
+ R
0
1
xh
,
c(x, t
k
) = e
rt
_
v(y, t
k+1
) f (y [ x)dy, t
k
T,
v(x, t
k
) = c(x, t
k
)1
x<h
+ R
0
1
xh
, t
k
T T
e
,
v(x, t
k
) = max
_
g(x), c(x, t
k
)
_
1
x<h
+ e
r(Tt
k
)
R
0
1
xh
, t
k
T
e
,
(24)
and
v(x, t
0
) = c(x, t
0
) = e
rt
_

v(y, t
1
) f (y [ x)dy,
where v(x, t
k
) = V(S
0
e
x
, t
k
) for any t
k
T, and h = log(H/S
0
).
Since each Lvy process is stationary and has independent increments, the condition density
f ( [ x) possesses the property:
f (y[x) = f (yx), x, y R,
where f (y) is the density of X
t
1
under the initial condition X
t
0
= x. Applying this property to
innite integrals c(x, t
k
) in (24) it becomes to
c(x, t
k
) = e
rt
_

v(y, t
k+1
) f (yx)dy = e
rt
_

v(x+z, t
k+1
) f (z)dz
for any t
k
T. Then, this integral can be rewritten as a convolution of v(x+z, t
k+1
) and the
function f (x), i.e.
c(x, t
k
) = e
rt
f (x) v(x, t
k+1
).
Thus we have
e
x
c(x, t
k
) = e
rt
_
e
x
f (x)
_
e
x
v(x, t
k+1
)
, (25)
where > 0 is chosen to improve the integrability. Now, the main idea of CONV method
is that, taking the Fourier transform in the both sides of (25) and applying the Property P3
(Convolution), the integral becomes
e
rt
T
_
e
x
c(x, t
k
)
(u) = T
__
e
x
f (x)
_
e
x
v(x, t
k+1
)
_
(u)
= T
_
e
x
f (x)
(u) T
_
e
x
v(x, t
k+1
)
(u).
Denote (z) is the characteristic function of the density f (x) on the complex plan C. Then, by
a simple calculation, we have
e
rt
T
_
e
x
c(x, t
k
)
(u) =
_
(ui)
_
T
_
e
x
v(x, t
k+1
)
(u).
Thus, taking the inverse Fourier transform we obtain
e
x
c(x, t
k
) = e
rt
1
2
_

e
iux
_
(ui)
_
_

e
iuy+y
v(y, t
k+1
)dydu. (26)
Denote
=
T
(u)
u
2
u=0
+
_
T
(u)
u
u=0
_
2
,
where
T
T
, and is a proportionality constant. According
to the suggestion from Lord et al (2008), we can take = 20 for the GBM model and = 40 for
other exponential Lvy models. Let N be a power of 2. We consider the grid points on x-axes:
x
j
=
_
j
1
2
N
_
x, j = 0, 1, . . . , N1,
where x = /N. Furthermore, we also consider the grid points for the numerical integrals
in (26):
u
j
= (j
1
2
N)u, y
j
= x
j
, j = 0, 1, . . . , N1,
where u = 2/. It is clear that these grids satisfy the Nyquist relation: uy = 2/N.
Now, for each t
k
T and each p = 0, 1, . . . , N1, approximating the integral in (26) with
composite trapezoidal rule and the second integral with left rectangle rule yields
c(x
p
, t
k
) e
x
p
rt
uy
2
N1
j=0
_
e
iu
j
x
p
_
(u
j
i)
_
N1
n=0
n
e
iu
j
y
n
+y
n
v(y
n
, t
k+1
)
_
where the weights
n
are chosen as
0
=
N1
=
1
2
and
n
= 1 for n = 1, . . . , N2. Noting
that u
0
=
1
2
Nu and x = y = 2/(Nu), we have
c(x
p
, t
k
) e
x
p
rt
e
iu
0
(y
0
x
0
)
(1)
p
N1
j=0
_
e
ijp
2
N
e
ij(y
0
x
0
)u
_
(u
j
i)
_
1
N
N1
n=0
e
ijn
2
N
(1)
n
n
e
y
n
v(y
n
, t
k+1
)
_
. (27)
Now, we can employ the FFT algorithm to calculate the summations in the right side of (27).
Once the integral c(x
p
, t
k
) is computed, we can determine the early-exercise price S
t
k
, t
k
T
e
,
by the procedure: for each t
k
T
e
, locate j
k
such that
c(x
j
k
, t
k
) g(x
j
k
) = 0, (28)
or,
_
c(x
j
k
, t
k
) g(x
j
k
)
__
c(x
j
k
+1
, t
k
) g(x
j
k
+1
)
_
0. (29)
In the case (28) set x
(t
k
) = x
j
k
, and in the case (29) set x
(t
k
) =
1
2
(x
j
k
+ x
j
k+1
). Then the
early-exercise price at every t
k
T
e
is given by S
t
k
= S
0
e
x
(t
k
)
. Ding et al (2011b) gave a detail
algorithm, which summarizes the above procedure, for pricing an up-and-out Bermudan
barrier option, and the corresponding numerical experiments.
4.2 The COS method for pricing Bermudan barrier options
Recently, Fang & Oosterlee (2009) extended their COS method to price discrete early-exercise
options under exponential Lvy models, and Fang & Oosterlee (2011) further considered such
pricing problems under Hestons model.
Assume that the price process of the underlying asset S
t
follows an exponential Lvy model,
and T and T
e
are the set of pre-specied monitored dates and the set of pre-specied exercise
dates, respectively, before the maturity T, which are dened by (22). In the following, we
apply the COS method to price the Bermudan barrier option which dened in preceding
subsection, whose payoff is given by
G(S
t
k
)1
S
t
k
<H
+ R
0
1
S
t
k
H
, t
k
T,
where H > K is the constant barrier and R
0
is the contractual rebate. Denote V(S, t
k
), t
k
T
e
,
the value of this Bermudan barrier option at time t
k
and the spot price S
t
k
= S. As in the
preceding subsection, with help of the risk-neutral valuation formula, this price process can
be computed recursively by the backward induction (23). In specialty, the initial price is given
by
V(S, t
0
) = C(S, t
0
) = E
_
e
rt
V(S, t
1
) [ S
t
0
= S
. (30)
Here S is the spot price of underlying asset, r > 0 is the interest rate. Let X
t
= log(S
t
/K) be
the logarithmof the underlying asset price S
t
over the strike price K, and denote x = log(S/K)
and h = log(H/K). Let f ( [ x) be the condition density of X
t
k+1
given X
t
k
= x for t
k
T. Set
g(x) =
_
K(e
x
1)
+
, for a call option,
K(1 e
x
)
+
, for a put option.
Then the backward induction (23) and the price formula (30) can be rewritten by
v(x, t
ML
) = g(x)1
x<h
+ R
0
1
xh
,
c(x, t
k
) = e
rt
_
v(y, t
k+1
) f (y [ x)dy, t
k
T,
v(x, t
k
) = c(x, t
k
))1
x<h
+ R
0
1
xh
, t
k
T T
e
,
v(x, t
k
) = max
_
g(x), c(x, t
k
)
_
1
x<h
+ e
r(Tt
k
)
R
0
1
xh
, t
k
T
e
,
(31)
and
v(x, t
0
) = c(x, t
0
) = e
rt
_

v(y, t
1
) f (y [ x)dy, (32)
where v(x, t
k
) = V(Ke
x
, t
k
) for any t
k
T.
We consider the innite integrals c(x, t
k
) in (31). Since f (y[x) decays to zero very quickly as
y we may choose two bounds a and b, which can be selected by (21), such that
c(x, t
k
) = e
rt
_
b
a
v(y, t
m+1
) f (y [ x)dy c(x, t
k
), t
k
T. (33)
without losing some signicant accuracy. Note that the density f (y [ x) has the following
Fourier-cosine expansion on [a, b]:
f (y [ x) =
2
b a
j=0
_
w
j
cos
_
j
y a
b a
_
_
b
a
f (u [ x) cos
_
j
u a
b a
_
du
_
,
where w
0
=
1
2
and w
j
= 1 for all j = 1, 2, 3, . . .. We substitute this expansion into the integral
(33) and then we can rewrite it by
c(x, t
k
) =
(b a)
2
e
rt

j=0
w
j
F
j
(x)V
j
(t
k+1
), (34)
where for each j = 1, 2, 3, . . .,
V
j
(t
k+1
) =
2
b a
_
b
a
v(y, t
k+1
) cos
_
j
y a
b a
_
dy, (35)
and
F
j
(x) =
2
b a
_
b
a
f (y [ x) cos
_
j
y a
b a
_
dy. (36)
Since the Fourier-cosine expansion has a high accuracy with a few terms, we can truncate the
innity series (34) and approximate c(x, t
k
) by leaving the rst N terms, i.e.,
c(x, t
k
) =
(b a)
2
e
rt
N1
j=0
w
j
F
j
(x)V
j
(t
k+1
) c(x, t
k
), (37)
for any t
k
T, where w
0
= w
N1
=
1
2
and w
j
= 1 for j = 1, . . . , N2. On the other hand, we
can represent each integral (36) by approximation as the following
F
j
(x) =
2
b a
Re
_
exp
_
i
ja
b a
_
_
b
a
f (u [ x) exp
_
i
ju
b a
_
du
_
2
b a
Re
_
exp
_
i
ja
b a
_
_

f (u [ x) exp
_
i
j
b a
u
_
du
_
.
Let (u; x) be the characteristic function of f ( [ x). Then, we can approximate each F
j
(x) by
F
j
(x)
2
b a
Re
_
exp
_
i
ja
b a
_

_
j
b a
; x
__
.
And so, we get the further numerical approximations of the integrals c(x, t
k
) in (37) by
c(x, t
k
) = e
rt
N1
j=0
w
j
Re
_
exp
_
i
ja
b a
_

_
j
b a
; x
__
V
j
(t
k+1
) c(x, t
j
). (38)
In special, the approximation of initial price v(x, t
0
) in (32) is given by
v(x, t
0
) = e
rt
N1
j=0
w
j
Re
_
exp
_
i
ja
b a
_

_
j
b a
; x
__
V
j
(t
1
). (39)
Meanwhile, since the characteristic function (u; x) possesses the property:
(u; x) = (u) e
iux
, u R,
where (u) = (u; 0), the approximations (38) and (39) can be simplied to
c(x, t
k
) = e
rt
N1
j=0
w
j
Re
_
exp
_
ij
x a
b a
_

_
j
b a
__
V
j
(t
k+1
), (40)
and the initial price of option
v(x, t
0
) = e
rt
N1
j=0
w
j
Re
_
exp
_
ij
x a
b a
_

_
j
b a
__
V
j
(t
1
). (41)
In order to use this approximate formulation we still need to compute the integrals V
j
(t
k
). For
convenience we introduce the following notions: for any a x
1
x
2
b,
C
j
(x
1
, x
2
; t
k
) =
2
b a
_
x
2
x
1
c(x, t
k
) cos
_
j
x a
b a
_
dx, j = 0, 1, . . . , N1,
and
D
j
(x
1
, x
2
) =
2R
0
b a
_
x
2
x
1
cos
_
j
x a
b a
_
dx, j = 0, 1, . . . , N1,
where c(x, t
ML
) = g(x), and c(x, t
k
), t
k
T, are given in (40).
We rst calculate V
j
(t
ML
), j = 0, 1, . . . , N1. We have
V
j
(t
ML
) =
_
C
j
(0, h; t
ML
) + D
j
(h, b), for a call option,
C
j
(a, 0; t
ML
) + D
j
(h, b), for a put option,
for all j = 0, 1, . . . , N1. Denote
j
(x
1
, x
2
) =
_
x
2
x
1
e
x
cos
_
j
x a
b a
_
dx and
k
(x
1
, x
2
) =
_
x
2
x
1
cos
_
j
x a
b a
_
dx,
for any a x
1
x
2
b and j = 0, . . . , N1. Then, by simple integration, these integrals
admit the following analytic solutions:
j
(x
1
, x
2
) =
1
1 + (
j
ba
)
2
_
cos
_
j
x
2
a
b a
_
e
x
2
cos
_
j
x
1
a
b a
_
e
x
1
+
j
b a
sin
_
j
x
2
a
b a
_
e
x
2
j
b a
sin
_
j
x
1
a
b a
_
e
x
1
_
,
and
j
(x
1
, x
2
) =
_
sin
_
j
x
2
a
b a
_
sin
_
j
x
1
a
b a
__
b a
j
,
for j = 0, . . . , N1, with
0
(x
1
, x
2
) = x
2
x
1
. Moreover, by a simple calculation, j =
0, . . . , N1, we have the following result:
D
j
(x
1
, x
2
) =
2R
0
b a
j
(x
1
, x
2
), (42)
C
j
(x
1
, x
2
; t
ML
) =
2
b a
K
_
j
(x
1
, x
2
)
j
(x
1
, x
2
)
_
, (43)
where is a parameter such that = 1 for a call option and = 1 for a put option.
And next, we consider to calculate the integrals V
j
(t
k
) for t
k
T T
e
. We have
V
j
(t
k
) = C
j
(a, h; t
k
) + e
r(Tt
k1
)
D
j
(h, b), j = 0, 1, . . . , N1.
Since the integral D
j
(x
1
, x
2
) has the analytic representation (45), we only need to calculate the
integral C
j
(x
1
, x
2
; t
k
). Fang & Oosterlee (2009) gave an efcient numerical algorithm which
approximates C
j
(x
1
, x
2
; t
k
) by using FFT method with the operation cost O(N log
2
(N)).
Finally, we consider to calculate the integrals V
k
(t
k
) for t
k
T
e
. It is clear that we should nd
the value v(x, t
k
) in the last equation in (31), or equivalently, to determine the early-exercise
point x
k
at each time t
k
, which is the point where the continuation value is equal to the payoff,
i.e., c(x
k
, t
k
) = g(x
k
). Let
h
k
(y) = c(y, t
k
) g(y), t
k
T
e
.
Then, the problem becomes to nd the root x
k
of each equation h
k
(y) = 0. Note that the
function c(y, t
k
), which is given in (40), is bounded and smooth, and the function g(y) is
smooth except for y = 0 and bounded in [a, b]. We can use the Newtons method or the
secant method to nd the root x
k
. Here if x
k
is not in the interval [a, b], we set it in the nearest
boundary point a or b. Once we nd the early-exercise point x
k
, t
k
T
e
, we have two different
cases for an up-and-out barrier option:
Case 1: x
k
< h, which means the early-exercise point doesnt hit the up barrier. Thus, We have
the authority to decide to execute the option now or reserve it to the next time point. So we
can split the integral that denes V
j
(t
k
) into three parts: [a, x
k
], (x
k
, h) and [h, b]. We have
V
j
(t
k
) =
_
C
j
(a, x
k
; t
k
) + C
j
(x
k
, h; t
ML
) + e
r(Tt
k1
)
D
j
C
j
(a, x
k
; t
ML
) + C
j
(x
k
, h; t
k
) + e
r(Tt
k1
)
D
j
(h, b), for a put option.
Case 2: x
k
h , which means the early-exercise point hits the up barrier. Thus, the option
integral can be split into two parts: [a, h) and [h, b]:
V
j
(t
k
) =
_
C
j
(a, h; t
k
) + e
r(Tt
k1
)
D
j
C
j
(a, h; t
ML
) + e
r(Tt
k1
)
D
j
(h, b), for a put option,
Ding et al (2011a) gave a detail algorithm, which summarizes the above procedure, for pricing
an up-and-out Bermudan barrier option, and the corresponding numerical experiments.
4.3 The fast Hilbert transform approach for pricing barrier options
Feng & Linetsky (2008) presented a new numerical method to price discretely monitored
barrier options under exponential Lvy models. Their method involves the relation with
Hilbert transform (Property P4) and the Sine expansion in Hardy spaces. They also gave
an efcient computational algorithm based on the fast Hilbert transform.
Let T = t
k
: k = 1, . . . , M be the set of pre-specied monitored dates, where
0 = t
0
< t
1
< < t
M
= T with t = t
k
t
k1
= T/M.
We consider a European-type barrier put option whose payoff at maturity T is given by
G = 1
S
t
1
>L
1
S
t
2
>L
1
S
t
M
>L
(K S
T
)
+
,
where K is the strike price and 0 < L < K is the lower barrier. We also assume that the
underlying asset price process is given by S
t
= Ke
X
t
with S
0
= S, where X
t
is a Lvy process
started at x = log(S/K). Denoting l = log(L/K), then, with help of the risk-neutral valuation
formula, the price of this option is given by
V(x, t
0
) = e
rT
E
_
1
X
t
1
>l
1
X
t
2
>l
1
X
t
M
>l
K(1 e
X
t
M
)
+
[ X
0
= x
,
which can be computed recursively by the following backward induction:
V(x, t
M
) = 1
x>l
K(1 e
x
)
+
,
V(x, t
k
) = e
rt
1
x>l
E
_
V(X
t
k+1
, t
k+1
) [ X
t
k
= x
, k = M1, . . . , 1,
V(x, t
0
) = e
rT
E
_
e
rt
V(X
t
1
, t
1
) [ X
t
0
= x
.
(44)
Since each Lvy process is stationary and has independent increments, for each k =
1, . . . , M1, we have
E
_
V(X
t
k+1
, t
k+1
) [ X
t
k
= x
= E
_
V(X
t
1
, t
k+1
) [ X
t
0
= x
.
Thus, letting P
t
1
v(x) = E
_
v(X
t
1
) [ X
t
0
= x
be the expectation operator, from the backward

induction (44) we get
v
M
(x) = 1
x>l
K(1 e
x
)
+
,
v
k
(x) = 1
x>l
P
t
1
v
k+1
(x), j = M1, . . . , 1,
v
0
(x) = P
t
1
v
1
(x),
(45)
where v
k
(x) = e
rt
V(x, t
k
). And hence, the problem now becomes to nd the function v
0
(x)
from the backward induction (45).
Note that 1
x>0
=
1
2
_
1 + sgn(x)
_
for all x R. Using the Property P4 (Relation with Hilbert
transform) we obtain
T(1
x>0
v)(u) =
1
2
_
Tv(u) + i1(Tv)(u)
_
. (46)
Let a R and T
a
be the transform operator: T
a
v(x) = v(x a). Then, we have
1
x>a
= T
a
1
x>0
=
1
2
_
1 +T
a
sgn(x)
_
,
and hence,
1
x>a
v(x) =
1
2
_
v(x) + v(x) T
a
sgn(x)
_
=
1
2
_
v(x) +T
a
_
sgn T
a
v
_
(x)
_
.
Taking the Fourier transform to both sides of this equation, we have
T
_
1
x>a
v
_
(u) =
1
2
Tv(u) +
1
2
T
_
T
a
_
sgn T
a
v
_
_
(u).
and using equation (46) we obtain
T
_
1
x>a
v
_
(u) =
1
2
Tv(u) +
1
2
ie
iau
1
_
e
iay
Tv(y)
_
(u). (47)
On other hand, noting that, for each k = 1, . . . , M1, the condition density of X
t
k+1
given
X
t
k
= x possesses the property:
f (y[x) = f (yx), x, y R,
where f (y) is the density of X
t
1
under the initial condition X
t
0
= x. The innite integrals
P
t
1
v
k+1
(x) in (45) becomes to
P
t
1
v
k+1
(x) =
_
+
v
k+1
(y) f (yx)dy =
_
+
v
k+1
(x+z) f (z)dz.
Then, this integral can be rewritten as a convolution of v
k+1
(x+z) and the function f (x), i.e.
P
t
1
v
k+1
(x) = f (x) v
k+1
(x).
Note that suppv
k
(x) (l, 0) for each k = M, . . . , 1 from the backward induction (45). We can
take the Fourier transform in the both sides of (45). Applying the Property P3 (Convolution)
we have
T
_
P
t
1
v
k
(x)
(u) = T
_
f (x) v
k+1
(x)
(u) = T
_
f (x)
(u) T
_
v
k+1
(x)
(u).
Denote (u) is the characteristic function of f (x) on the complex plane C. Then, by a simple
calculation, we have
T
_
P
t
1
v
k
(x)
(u) = (u) Tv
k+1
(u). (48)
Thus, using the formulas (47) and (48) we obtain
v
M
(u) =
K(1e
iul
)
iu

K(1e
(1+iu)l
)
1+iu
,
v
k
(u) =
1
2
(u) v
k+1
(u) +
1
2
ie
iul
1
_
e
iyl
(y) v
k+1
(y)
_
(u),
k = M1, . . . , 1,
v
0
(u) = (u) v
1
(u).
(49)
Here v
k
(u) = Tv
k
(u) for each k. Applying the truncated Sinc approximation, Feng &Linetsky
(2008) obtained the discretization of v
k
(u):
v
k
(nh) =
1
2
(nh) v
k+1
(nh)
+
ie
inhl
2
N
j=N,j,=n
e
ijhl
(jh) v
k+1
(jh)
1 (1)
nj
n j
, (50)
for n = N, . . . , N and k = M1, . . . , 1, where N is a positive integer and h is the
discretization step size. Then, the function v
0
(x) can be computed by the discretised inversion
Fourier transform:
v
0
(x) =
1
2
N
j=N
e
ijhx
(jh) v
1
(jh)h.
Furthermore, Feng & Linetsky (2008) showed that the computation (50) involves a Toeplitz
matrix-vector multiplication, which can be accomplished in O(N log
2
N) operations using the
FFT technique. They also referred the corresponding algorithm of computing the discrete
Hilbert transform via the FFT as the Fast Hilbert Transform.
5. References
Barndorff-Nielsen, O. (1997). Normal inverse Gaussian distributions and stochastic volatility
modelling, Scandinavian Journal of Statistics, Vol. 24, pp. 113.
Bates, D. S. (1996). Jumps and stochastic volatility: exchange rate processes implicit in
Deutsche mark options, Rev. Financ. Stud., Vol. 9, pp. 69107.
Black, F. &Scholes, M. (1973). The pricing of options and corporate liabilities, Journal of Political
Economy, Vol. 81 (No. 3), pp. 637654.
Borak, S., Detlefsen, K. & Hrdle, W. (2005). FFT based option pricing, SFB 649 Discussion
paper 2005-011. http://sfb649.wiwi.hu-berlin.de
Carr, P., Geman, H., Madan, D. & Yor, M. (2002). The ne structure of asset returns: An
empirical investigation. Journal of Business, Vol. 75, pp. 305332.
Carr, P. & Madan, D. (1999). Option valuation using the fast Fourier transform, Journal of
Computational Finance, Vol. 2 pp. 6173.
Carr, P. & Wu, L. (2003). The nite moment log stable process and option pricing, Journal of
Finance, Vol. 58, pp. 753777.
Cont, R. & Tankov, P. (2003). Financial Modelling with Jump Processes, Chapman & Hall/ CRC
Press.
Dempster, M. A. & Hong, S. S. G. (2002). Spread option valuation and the fast Fourier
transform, Mathematical FinanceBachelier Congress 2000, Spring, pp. 203220.
Ding, D & U, S. C. (2010). An accurate and stable FFT-based method for pricing options under
exp-Levy processes, The Proceedings of ISCM II - EPMSC XII, American Institute of
Physics, pp. 741746.
Ding, D & U, S. C. (2011). Efcient option pricing methods based on Fourier series expansions,
Journal of Mathematical Research and Exposition, Vol. 31, pp. 1222.
Ding, D., Huang, H. & Zhao, J. (2011a). An efcient algorithm for Bermudan barrier option
pricing, Working paper, University of Macau.
Ding, D., Weng, Z. &Zhao, J. (2011b). Pricing Bermudan barrier options via a fast and accurate
FFT-based Method, to appear in the Proceedings of CiSE 2011, IEEE.
Dufe, D., Pan, J. & Singleton, K. (2000). Transform analysis and asset pricing for afne
jump-diffusions, Econometrica, Vol. 68, pp. 13431376.
Dufresne, D., Garrido, J. & Morales, M. (2009). Fourier inversion formulas in option pricing
and insurance, Methodol. Comput. Appl. Probab., Vol. 11, pp. 359383.
Eberlein, E., Glau, K., & Papapantoleon, A. (2010). Analysis of Fourier transform valuation
formulas and application, Applied Mathematical Finance, Vol. 17, pp. 211240.
Fang, F. & Oosterlee, C. W. (2008). A novel pricing method for European option based on
Fourier-cosine series expansions, SIAM J. Sci. Comput., Vol. 31, pp. 826848.
Fang, F. & Oosterlee, C. W. (2009). Pricing early-exercise and discrete barrier options by
Fourier-cosine series expansions, Numer. Math., Vol. 114, pp. 2762.
Fang, F. & Oosterlee, C. W. (2011). A Fourier-based valuation method for Bermudan and
barrier options under Hestons model, SIAM J. Financial Math., Vol. 2, pp. 439463.
Feng, L. & Linetsky, V. (2008). Pricing discretely monitored barrier options and defaultable
bonds in Lvy process models: a fast Hilbert transform approach. Mathematical
Finance, Vol. 18, pp. 337384.
Gurland, J. (1948). Inversion formula for the distribution of ratios, Ann. Math. Statist., Vol. 19,
pp. 228237.
Heston, S. (1993). A closed-form solution for options with stochastic volatility with
applications to bond and currency options, Rev. Financ. Stud., Vol. 6, pp. 327343.
Hurd, T. & Zhou, Z. (2010). A Fourier transform method for spread option pricing, SIAM J.
Finan. Math., Vol. 1, pp. 142157.
Lee, R. (2004). Option pricing by transform methods: Extensions, unication, and error
control, Journal of Computational Finance, Vol. 7, pp. 5186.
Lewis, A. (2001). A simple option formula for general jump-diffusion and other exponential
Lvy processes. http://www.optioncity.net
Lord, R., Fang, F., Bervoets, F. &Oosterlee, C. W. (2008). Afast and accurate FFT-based method
for pricing early-exercise options under Lvy processes. SIAM J. Sci. Comput., Vol. 30,
pp. 16781705.
Kou, S. G. (2002). A jump-diffusion model for option pricing, Management Science, Vol. 48, pp.
1086-1101.
Kwok, Y. K., Leung, K. S. & Wong, H. Y. (2010). Efcient options pricing using the fast Fourier
transform. http://ssrn.com/abstract=1534544
Madan, D., Carr, P. & Chang, E. (1998). The variance gamma process and option pricing,
European Financial Review, Vol. 2, pp. 79105.
Merton, R. (1973). Theory of rational option pricing, Bell Journal of Economics and Management
Science , Vol. 4, pp. 141183.
Merton, R. (1976). Option pricing when underlying stock returns are discontinuous. J. Financ.
Econ., Vol. 3, pp. 125144.
Schmelzle, M. (2010). Option Pricing Formulae using Fourier Transform: Theory
and Application. http://pfadintegral.com/articles/option-pricing-formulae-using-
fourier-transform/
Shephard, N. G. (1991). From characteristic function to distribution function: a simple
framework for the theory, Econometric Theory, Vol. 7, pp. 519529.
Stein, E. & Stein, F. (1991). Stock price distribution with stochastic volatility: An analytic
approach, Rev. Financ. Stud., Vol. 4, pp. 727752.
0
Hilbert Transform and Applications
Yi-Wen Liu
National Tsing Hua University
Taiwan
1. Introduction
Hilbert transformnds a companion function y(t) for a real function x(t) so that z(t) = x(t) +
iy(t) can be analytically extended from the real line t Rto upper half of the complex plane.
In the eld of signal processing, Hilbert transform can be computed in a few steps:
First, calculate the Fourier transform of the given signal x(t). Second, reject the negative
frequencies. Finally, calculate the inverse Fourier transform, and the result will be a
complex-valued signal where the real and the imaginary parts form a Hilbert-transform pair.
When x(t) is narrow-banded, |z(t)| can be regarded as a slow-varying envelope of x(t)
while the phase derivative
t
[tan
1
(y/x)] is an instantaneous frequency. Thus, Hilbert
transformcan be interpreted as a way to represent a narrow-band signal in terms of amplitude
and frequency modulation. The transform is therefore useful for diverse purposes such as
latency analysis in neuro-physiological signals (Recio-Spinoso et al., 2011; van Drongelen,
2007), design of bizarre stimuli for psychoacoustic experiments (Smith et al., 2002), speech
data compression for communication (Potamianos & Maragos, 1994), regularization of
convergence problems in multi-channel acoustic echo cancellation (Liu & Smith, 2002), and
signal processing for auditory prostheses (Nie et al., 2006).
The rest of this review chapter is organized as follows: Sec. 2 reviews the mathematical
denition of Hilbert transform and various ways to calculate it. Secs. 3 and 4 review
applications of Hilbert transform in two major areas: Signal processing and system
identication. The chapter concludes with remarks on the historical development of Hilbert
transform in Sec. 6.
2. Mathematical foundations of Hilbert transform
The desire to construct the Hilbert transform stemmed from this simple quest: Given a
real-valued function f : R R, can we nd an imaginary part ig such that f
c
= f + ig
can be analytically extended? For example, if f (x) = cos(x), then by inspection we can nd
g(x) = sin(x) such that f
c
(x) = f + ig = exp(ix). This function can obviously be extended
analytically to the entire complex plane by replacing the real variable x with the complex
variable z in the expression; the result is f
ext
(z) = exp(iz) and we have
Re{f
ext
(z)}|
z=x
= f (x), (1)
12
which states that real part of the extended function is equal to the original given function f (x)
on the real line. The companion function g(x) is called the Hilbert transform of f (x).
This simple example brings forth a few questions. First, is analytic extension always possible
for reasonably smooth f (x)? If so, is the companion function g(x) unique, in the sense that
Eq. (1) would hold? The answer to the second question is a denite No, because any f
c
(x) =
f (x) + i{g(x) + g
0
} also satises Eq. (1), where g
0
is an arbitrary constant. Therefore, the
original simple quest needs to be rened as follows.
2.1 Hilbert transform as a boundary-value problem
To establish the uniqueness of the companion function, we rst note that any analytic
function f
ext
(z) = f
R
(z) + i f
I
(z) dened on the complex plane z = x + iy must satisfy
Cauchy-Riemann equations,
f
R
x
=
f
I
y
,
f
I
x
=
f
R
y
.
Consequently, both f
R
and f
I
satisfy Laplaces equation,
2
F
R
x
2
+

2
F
R
y
2
= 0, (2)
2
F
I
x
2
+

2
F
I
y
2
= 0 (3)
over the region where f
ext
(z) is analytic. Conventionally, by requiring f
ext
(z) to be analytic in
the upper half-plane, the quest of nding the Hilbert transform for any given function f (x)
can be formulated as a boundary value problem (Scott, 1970). By specifying the boundary
conditions that
f
R
(x, 0) = f (x), and that
f
R
(x, y) = 0 as x or y ,
f
R
(x, y) can be uniquely determined by solving Laplaces equation (Eq. 2) in the upper half
plane. Then, through Cauchy-Riemann equations, f
I
(x, y) can be calculated in the entire
upper half plane and in particular on its boundary the x-axis (Scott, 1970). Thus g(x) =
F
I
(x, 0) is the Hilbert transform of the given function f (x).
2.2 Calculation through improper integrals
The above formulation of Hilbert transformas a boundary-value problemis hardly mentioned
in recent texts [except as an exercise problem in Oppenheim & Schafer (2010)].
1
: Instead,
1
The boundary-value-problem formulation is missing for a reason it does not tell us how to compute
the Hilbert transform.
Hilbert Transform and Applications 3
Hilbert transform is commonly introduced and dened through an improper integral [e.g.,
(Hahn, 1996)]:
g(x) =
1
f (u)
1
x u
du. (4)
Here, note that the convolution kernel function h(x) = 1/x is singular at x = 0. Therefore,
the integral in Eq. 4 is improper in the sense of Cauchys principal value:
g(x) = lim
0

x+
f (u) h(x u)du. (5)

To be convinced that Eq. 4 indeed produces the Hilbert transform, we need to think about
the effects of Hilbert transform in the frequency domain. First, for any frequency k, note that
the Hilbert transform of f
k
(x) = cos(kx) is g
k
(x) = sin(kx). So, we can understand Hilbert
transform as a phase shifter which gives every sinusoidal function 90 degrees of phase shift.
Therefore, in the frequency domain, we have
G(k) = F(k) (i sgn(k)) , (6)
where G(k) and F(k) are the Fourier transform of g(x) and f (x), respectively, and sgn(x) is
the sign function (i.e., sgn(k) = 1 if k > 0 and sgn(k) = 1 if k < 0.) Therefore, if we think
of H(k) = i sgn(k) as the transfer function of a phase-shift kernel h(x), the kernel can be
written as the inverse Fourier transform of the transfer function; that is,
h(x) =
1
2
H(k)e
ikx
dk. (7)
Note that H(k) = i for k > 0 and H(k) = i for k < 0. Therefore, H(k)s rst derivative with
respect to k is
H
k
= 2i(k), (8)
where (k) is the Dirac delta function. Since the operator /k in the frequency domain
corresponds to multiplication by ix in the space domain, we can take the inverse Fourier
transform on both sides of Eq. 8 and obtain the following,
ix h(x) = 2i
1
2
(k)e
ikx
dk
=
i
. (9)
Dividing both sides by ix, we conclude that the convolution kernel h(x) = 1/x.
2.3 The notion of Hilbert transform pairs
The phase-shifter interpretation of Hilbert transform leads to the fact that if f (x)s Hilbert
transform is g(x), then g(x)s Hilbert transform is f (x); in this sense, f (x) and g(x) form a
Hilbert transform pair.
This symmetric property can be understood as follows. Note that the H
2
(k) = 1 for all k
since H(k) = i. This means that if we take the Hilbert transform twice, the result would be
the original function with a negative sign.
293 Hilbert Transform and Applications
This makes sense because Hilbert transform introduces a 90-degree phase shift to all simple
harmonics. Therefore, Hilbert transform repeated twice introduces a 180-degree phase shift
to all simple harmonics, which means multiplication of the original function by 1.
A table of commonly used Hilbert transform pairs can be found in the Appendix of Hahn
(1996) for applications in signal processing. A thorough 80-page table of Hilbert transform
pairs can be found in the Appendix of King (2009b) and transform pairs are also plotted in a
20-page atlas.
2.4 The convolution kernel h(x) as the Hilbert transform of (x)
We now introduce another way to derive the convolution kernel of Hilbert transform. To
begin, note that Eq. 5 essentially states that Hilbert transform is a ltering process which is
characterized by its impulse response h(x). Therefore, h(x) must be regarded as the Hilbert
transform of the impulse function (x). Then, it is of our interest to check that
f
c
(x) = (x) + ih(x)
can be regarded as an analytic function in the sense of Eq. 1. To see it, consider a family of
complex analytic functions f (z) = i/(z + i) parametrized by a variable > 0. Since the
only singularity of f (z) is at z = i, f (z) is analytic in the entire upper half plane. Therefore,
the real part and imaginary part of f (z) form a Hilbert transform pair on the real line x R.
With a little algebra, the real and imaginary parts can be written as
f (x) =
i
(x + i)
= f
R
(x) + i f
I
(x), (10)
where
f
R
(x) =

(x
2
+
2
)
and
f
I
(x) =
x
(x
2
+
2
)
form a Hilbert transform pair for any > 0.
Now we let approach zero and observe f
R
(x) and f
I
(x). Note that

f
R
(x)dx = 1
regardless of the value of , and that f
R
(0) = 1/ approaches innity as 0. So we
can claim that
lim
0
f
R
(x) = (x). (11)
Meanwhile, it is trivial that
lim
0
f
I
(x) = 1/x. (12)
From the arguments above, we can be convinced that the Hilbert transform of (x) is indeed
1/x, for f
c
(x) = (x) +i (1/x) is equal to the limit function of f (x) as 0 (Scott, 1970).
3. Applications in signal processing
Signal processing is nowadays conveniently conducted in the digital domain. The rst step of
signal digitization involves sampling an analog signal x(t) at a constant rate f
s
= 1/T where
T is the sampling period. In this section, we denote the sampled waveform as x[n] = x(nT),
using the square brackets [] to indicate that the signal is sampled in discrete time. So the
discrete-time Fourier transform (DTFT) is dened as follows:
2
X(j) =
n=
x[n]e
jn
.
Note that X(j) is periodic at every 2 in the frequency domain.
3
Next, we showhowHilbert
transform can be dened in discrete time.
3.1 The discrete-time Hilbert transform and Hilbert transformers
Recall that the Hilbert transform introduce a 90-degree phase shift to all sinusoidal
components. In the discrete-time periodic-frequency domain, the transfer function of Hilbert
transform is specied as follows,
H(j) =
j, 0 < <
j, < < 0
(13)
The convolution kernel for H(j) can be calculated through inverse Fourier transform
(Oppenheim & Schafer, 2010):
h[n] =
1
2
H(j)e
jn
d (14)
=
sin
2
(n)
n
, n = 0
0, n = 0
(15)
Note that h[n] has a innite support from n = to . In practice, the entire function can
not be stored digitally. To circumvent this difculty, we now discuss two major methods for
calculating the discrete-time Hilbert transform.
3.1.1 The MATLAB approach
The universally popular scientic-computing software MATLAB (MathWorks, Natick,
Massachusetts, USA) has a hilbert() function that computes the so-called discrete-time
analytic signal X = Xr + i
*
Xi such that Xi is the Hilbert transform of Xr.
4
MATLABs
implementation of the hilbert() function takes advantage of the fast Fourier transform
(FFT). Essentially, the hilbert() function completes the calculation in three steps:
Do the FFT of Xr.
Set the elements in FFT which correspond to frequency < < 0 to zero.
Do the inverse FFT.
2
We now switch to the electrical-engineering convention of using j to refer to
1.
3
Hereafter, we use capital letters X, Y, Z, H, ... to denote spectrums in the frequency domain, and
lowercase letters x, y, z, h for signals in the time domain.
4
This is what MATLABs help le shows.
That the three steps above can work is a consequence of the fact that, if x
i
[n] is the discrete
Hilbert transform of x
r
[n], the Fourier transform of x
r
[n] + jx
i
[n] vanishes for all negative
frequencies < < 0. This can be veried by inspecting the denition given in Eq. 13.
Remarks: Perhaps because of MATLABs popularity and easiness to use, many have allegedly
mis-regard the returned vector X=hilbert(Xr) as the Hilbert transform of Xr. One must be
aware of these deviations from the conventional denition to avoid unnecessary confusion.
3.1.2 Hilbert transform as a lter design problem
Thanks to FFT, MATLABs implementation of Hilbert transform is very efcient. However,
it should be used with caution note that the Hilbert transform dened in Eq. 13 has a
discontinuity at = 0. Consequently (due to Gibbs phenomenon), the convolution kernel
h[n] in Eq. 15 has an innite support in time. As a result, when implemented through FFT, the
Hilbert transform kernel wraps around itself and time-domain aliasing comes in (Oppenheim
&Schafer, 2010). The time-domain aliasing could be perceived as artifacts in applications such
as audio and video signal processing.
To avoid time-domain aliasing, one can formulate discrete Hilbert transform as a lter-design
problem. The ideal transfer function is specied by Eq. 13, and there are standard techniques
to design nite impulse response (FIR) or innite impulse response (IIR) lters that appraoch
the ideal transfer function. For example, one can truncate the ideal impulse response by
multiplying it with a window function w[n] which has a nite support (lets say from n = N
to N). Then the resulting function w[n]h[n] yields an approximate magnitude response
H
w
(j) that has a smooth transition between negative and positive frequencies as well as
ripples in both regions. The height of the ripples can be reduced by selecting the window
function wisely, while the transition bandwidth is inversely proportional to the window
length. Interested readers can refer to Chapter 7 and 12 of Oppenheim & Schafer (2010).
The FIR or IIR lters designed to approximate Hilbert transformare called Hilbert transformers.
Next, we discuss applications of Hilbert transformers in communication and in biomedical
engineering.
3.2 Sampling of bandpass signals for communication
An important application of Hilbert transformers is in sampling bandpass signals.
5
To explain
this, let us assume that a bandpass signal s(t) is has a region of support f
c
f f
c
+f in
the frequency domain, where f = 0.2f
c
. Based on Nyquists theorem, the sampling rate
needs to be at least two times the highest frequency, or 2.4f
c
, to avoid frequency-domain
aliasing. However, the bandwidth of this signal is really f = 0.2f
c
, so f
s
= 2.4f
c
is in fact an
oversampling.
To take advantage of the narrow bandwidth, we initially need to sample at 2.4f
c
to obtained
an oversampled signal s
r
[n] = s(nT), where T = 1/f
s
. Then, we can use a Hilbert transformer
to obtain s
i
[n] such that the Fourier transform of s[n] = s
r
[n] + js
i
[n] has no components at
negative frequencies < < 0. Now, S(j)s region of support is (5/6, ). Therefore,
we can down sample s[n] by a factor of 6 and there would be no frequency-domain aliasing.
5
This part is adapted from Sec. 12.4.3 of Oppenheim & Schafer (2010)
Transmitting s
d
[n] = s[6n] is more efcient than transmitting s
r
[n] because the sampling rate
is lowered.
To reconstruct s
r
[n] from s
d
[n], we can do the following:
Expand s
d
[n] by a factor of 6; i.e., construct s
e
[n] =
s
d
[n/6], if n = 0, 6, 12, ...
0, otherwise.
Filter s
e
[n] with the passband of (5/6, ).
Take the real part, and the result is a reconstructed copy of s
r
[n].
In practice, since all Hilbert bandpass lters are not ideal, one needs to consider sampling
at slightly higher than 2.4f
c
. That would protect the passband from ripple interference both
during downsampling and during signal reconstruction.
3.3 AM-FM decomposition for auditory prostheses
A cochlear implant device consists of up to tens of electrodes which are inserted as an array to
the cochlea to stimulate auditory nerves by electrical currents (Zeng et al., 2008). Each channel
(electrode) represents an acoustic frequency band; the amount of currents sent to an electrode
should faithfully reect how acoustic energy entering through the ear microphone varies in
time in the corresponding frequency band. Typically, acoustic waveforms are processed with
a bank of lters and the resulting envelopes control the electrical currents sent to individual
electrodes.
In this application scenario, a Hilbert transformer has been found useful for envelope
extraction. This can be understood by noting that the Hilbert transformproduces a sin(t) for
every cos(t). For a narrow-band signal s
R
(t), we can factor it as a product of slow-varying
envelope A(t) and fast-varying ne structure f (t):
s
R
(t) = A(t) f (t) = A(t) cos[(t)]
where d(t)/dt can be regarded as an instantaneous frequency of the signal. It can be shown
that, if A(t) varies sufciently slowly, the Hilbert transform produces approximately s
I
(t) =
A(t) sin[(t)]. Then, the envelope A(t) can simply be estimated by taking the root of sum of
squares:
A(t)
s
2
R
(t) + s
2
I
(t). (16)
Clinically, encoding the currents based on A(t) provides clear speech perception for
cochlear-implant users (Nie et al., 2006). Moreover, auditory pitch information can be
extracted by taking the time-derivation of (t), which can be determined by the relative phase
between s
I
and s
R
:
(t) = tan
1
s
I
(t)
s
R
(t)
.
Pitch is arguably the utmost important feature for music perception as well as understanding
tonal languages. Here, we see Hilbert transform can serve as an efcient computation tool to
extract it.
4. Applications in system identication
Hilbert transform relates the real part and the imaginary part of the transfer function of
any physically viable linear time-invariant system. By physical viability we mean a system
should be stable and causal. Stability requires the systems to produce bounded output if
the input is bounded [so called the bounded-input bounded-output criterion, (Oppenheim &
Schafer, 2010)]. Causality prohibits the system from producing responses before any stimulus
comes in. Denote the impulse response as h(t) and its Laplace transform as H(s). The above
conditions requires that
h(t) = 0 for all t < 0 (causality)
All sigularities of H(s) are located in the left half-plane (stability).
The two conditions above ensure that H(s) converges and is analytic in the entire right
half-plane, and in particular on the imaginary axis s = j. Therefore, the real and imaginary
part of H(j) = H
R
() + jH
I
() are inter-dependent in term of the Kramers-Krnig relations
(King, 2009b):
H
I
() =
1
H
R
(u)du
u
(17)
H
R
() =
1
H
I
(u)du
u
(18)
which is basically Hilbert transform in its time-frequency dual form.
The Kramers-Krnig relations govern how physical viable transfer functions can vary in
frequency. For instance, the real and imaginary part of an electromagnetic wave propagation
function denes the attenuation and the wavenumber per unit length as a function of
frequency. The frequency dependence of both functions is referred to as dispersion, for in optics
it describes how red light travels faster than violet light in water so we behold the beauty
of rainbows after raining. It is now intriguing to realize that the attenuation and velocity of
light are two inter-dependent functions of frequency. More physics-inclined readers can refer
to Kings thorough discussion on dispersion relations in electrodynamics and optics (King,
2009b).
To a certain extent, the concept that the real and the imaginary parts are inter-dependent
similarly applies to the magnitude and phase of transfer functions of a physically viable
system. Note that any transfer function H(j) can be decomposed logarithmically into
magnitude and phase:
log H(j) = log |H(j)| + jH(j).
This shows that the log-magnitude and the phase are real and imaginary parts of the
log-spectrum, respectively. It might appear that they must satisfy the Kramers-Krnig
relations. Unfortunately, this is a wishful thinking since apparently

H(j) =
exp(j)H(j), where is a constant, would have the same magnitude as H(j) but a
different phase response.
6
It turns out that, for any given magnitude response, the uniqueness of phase response can be
established if the transfer function satises a minimum-phase criterion; the criterion requires
6
in fact,

h(t) = h(t ).
that all zeros and poles of the transfer function H(s) to be located in the left-half plane. This
criterion ensures that all the singularities of log H(s) are located in the left-half plane so the
real and imaginary parts of log H(s) become a Hilbert transform pair. Otherwise, any transfer
function can be uniquely factorized as a product of a minimum-phase function M(j) and
an all-pass function P(j). It is noteworthy that the system whose transfer function is M(j)
has the minimal energy delay among all linear time-invariant systems of the same magnitude
response. Further readings are recommended in Chapter 5 and 12 of Oppenheim & Schafer
(2010).
5. Concluding remarks and historical developments
In this chapter, we rst presented Hilbert transform as an analytic extension problem. Hilbert
transform uniquely exists due to Cauchy-Riemann equations. We then reviewed several
different ways to calculate Hilbert transform. A few important points stand out: First, Hilbert
transform can be regarded as a 90-degree phase shifter. Secondly, the real part and imaginary
part of a physically viable transfer function must satisfy Kramers-Krnig relations, which is
the Hilbert transform applied in time-frequency duality. A good reference on these topics can
be found in Oppenheim & Schafer (2010).
The construction of Hilbert transform pairs through Cauchy-Riemann equations in Sec. 2.1
was found in the appendices of an old text on microwave electronics (Scott, 1970). The
original formulation was stated in terms of Kramers-Krnig relation, and in this chapter that
formulation is adapted so the signal is dened on the real line instead of the frequency axis
j.
As a matter of fact, the denition of Hilbert transformwas not given by David Hilbert himself.
The name Hilbert transform was rst given by the British mathematician G. H. Hardy in
honor of Hilberts pioneering work on integral equations (King, 2009a). Hardys early work
established mathematical rigor of the transform(Hardy, 1932), which we nowapply in various
areas such as physiology and telecommnication. Areviewof these contributions frombrilliant
mathematicians reminds us that the transform really is a heritage from the 20th century.
Nevertheless, it is also amusing that nowadays we calculate Hilbert transform with super
fast computers, which might never had been envisioned by 20th-century pioneers, including
Hilbert, Hardy, Scott, or even Oppenheim.
6. Acknowledgement
This work is supported by Taiwans National Science Council through research grant
No. 100-2221-E-007-011.
7. References
Hahn, S. L. (1996). Hilbert Transforms in Signal Processing, Artech House, Inc., Norwood, MA,
USA.
Hardy, G. H. (1932). On hilbert transforms, Quart. J. Math. (Oxford) 3: 102112.
King, F. W. (2009a). Hilbert Transforms, Vol. 1, Cambridge University Press, Cambridge, UK.
King, F. W. (2009b). Hilbert Transforms, Vol. 2, Cambridge University Press, Cambridge, UK.
Liu, Y.-W. & Smith, J. O. (2002). Perceptually similar orthogonal sounds and applications to
multichannel acoustic echo canceling, Proc. Audio Eng. Soc. 22nd Int. Conf., Espoo,
Finland.
Nie, K., Barco, A. & Zeng, F.-G. (2006). Spectral and temporal cues in cochlear implant speech
perception, Ear Hear. 27: 208217.
Oppenheim, A. V. & Schafer, R. W. (2010). Discrete-Time Signal Processing, 3rd Edition, Pearson,
Boston.
Potamianos, A. & Maragos, P. (1994). A comparison of the energy operator and the hilbert
transform approach to signal and speech demodulation, Signal Processing 37(1): 95
120.
Recio-Spinoso, A., Fan, Y.-H. & Ruggero, M. (2011). Basilar-membrane responses to
broadband noise modeled using linear lters with rational transfer functions,
Biomedical Engineering, IEEE Transactions on 58(5): 1456 1465.
Scott, A. (1970). Active and Nonlinear Wave Propagation in Electronics, Wiley-Interscience, New
York.
Smith, Z. M., Delgutte, B. & Oxenham, A. J. (2002). Chimaeric sounds reveal dichotomies in
auditory perception, Nature 416: 8790.
van Drongelen, W. (2007). Signal Processing for Neuroscientists: Introduction to the analysis of
physiological signals, Academic Press, London.
Zeng, F.-G., Rebscher, S., Harrison, W., Sun, X. & Feng, H. (2008). Cochlear implants: System
design, integration, and evaluation, IEEE Rev. Biomed. Eng. 1: 115142.

Fourier Transform Applications

Uploaded by

Document Information

Original Description:

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Fourier Transform Applications

Uploaded by

Copyright:

Available Formats

FOURIER TRANSFORM

J(r, t). (1)

) in this equation for a specic

), including the broadside ( = 0

] takes the form

squared term in this equation is due to the quadratic phase

). The RTSAsystemis calibrated employing a linear frequency ramp.

(t) is the second

the incomplete Gamma functions,

can be obtained efficiently via, for example, a

w belongs to the boundary of S . Therefore,

w is our sought solution to the NICCB

for NECCB is given by:

is the same as NICCB, but the solution in the bound of

is the true solution to NECCB.

to NECCB, but the

is the true solution to NECCB.

is the true solution to

is the true solution to

. Hence, the loading level has a great impact on

is the jammer powers of the k-1-th iteration, K is the iterative coefficient.

= w a is the pattern of the k-1-th iteration, therein w is the relative weight

in the polynomial, and the solving process will

= R X X is the estimating formula of R. Thus (4.25) is established. When the

= R X X , and transform the received

R , and get the signal

E , and the source number N

, the intensity of gyro-synchrotron

(Dulk, 1985). According to the estimations in

1), where B/B

, given by equation (Dulk, 1985)

and absorption coefcient

. For the last, Dulk (1985) provides the following

, where is an integer greater than zero

, where / , is an integer greater than zero.

is a message vector of length p

is the Fourier variable corresponding to x

()v in (1.8), we showed the following pointwise estimate with respect to R

), and C is a positive constant.

norm with respect to x

). For the above results, we refer to the reeder

is the Fourier variable corresponding to x

. For a nonnegative integer s, we denote by H

) denotes the space of L

(u) < 0 for u [u

(u) < 0 for u [u

is the Fourier variable corresponding to x

norm with respect to x

. Then it holds that (3.4)(3.6) with = 1.

z and take the real part, obtaining

( 0) and integrate over [0, t].

and t 0, where the norms | |

are with respect to x

is a constant depending on . We apply this inequality to the term on

( 0) and integrate over [0, t]. This yields

(0 1) and the Hlder inequality, we see that

be the expectation operator, from the backward

f (u) h(x u)du. (5)

You might also like